W whatember what ember is up to · Nov 2026
08 · DeOS, MiniDregg, NanoDregg, DelveTalk, OBend Language & The Breadstuffs Diaspora

From the 2.89M-LOC breadstuffs Monorepo to Verified minidregg, nanodregg, delvetalk2 & Objective Bend (obend)

Complete primary-source archaeology across the 100+ crates and 10,639 commits of ~/dev/breadstuffs, the Objective Bend (obend) language, Stage-B sealed Cathedral (597 Tier-1 modules) & Amadio–Cardelli equi-recursive bisimulation, and the 315 worktree lanes of minidregg-lanes, 67 recovered lanes of claudesplosion, 27 scouts of nanodregg-scout, 66 lanes of delvetalk2-lanes, drorb, and DreggCloud.

Dossier 09 · DeOS, MiniDregg, NanoDregg, DelveTalk, Spween, Preoscript & OBend Deep Dive (64 KB)
Working field notes & source index below, with theorem anchors, code pointers, git history, and open questions.

Strand 09 — DeOS, Mini, Nano, DelveTalk, Spween, preoscript and the Objective Bend convergence

Primary-source dossier, written 2026-11-02 from the repositories, git logs, memory files and review corpora named in each section. Every number below was read from a file or a git/wc command on 2026-11-02 unless a dated document is cited for it. Hellas is not part of this strand; where the memory record mentions it, it appears only as the boundary Ember drew ("that is a hellas thing, not a dregg thing", 10-01).


0. How to read this, and four corrections to the brief

0.1 Source ledger

sourcepathwhat it settles
Mini memory (80 files)~/.claude/projects/-Users-ember-dev-minidregg/memory/mission, design laws, every Oct ruling (retained mode, never-slower, board/seats, demo goal)
breadstuffs memory (330 files)~/.claude/projects/-Users-ember-dev-breadstuffs/memory/09-29 decisions, 09-30 cutover, 10-08 drain-and-freeze, nano rebuild + wind-down, DeOS/firmament epochs
DelveTalk memory (12)~/.claude/projects/-Users-ember-dev-delvetalk/memory/the foundation takeover, "horseshit", posting-first, tulip privacy
obend memory (16)~/.claude/projects/-Users-ember-dev-obend/memory/modularity-first, no naive harness, ground-yourself-before-fanout
DreggNet memory (30)~/.claude/projects/-Users-ember-dev-DreggNet/memory/abandon-vs-operate, converged architecture, firmament/seL4, drorb
claudesplosion memory (4)~/.claude/projects/-Users-ember-dev-claudesplosion/memory/staffing (Sonnet fan-out, Opus review), hbox push route
review corpus~/dev/claudesplosion/{00..23,mini-1..6,99-SYNTHESIS,bakeoff/VERDICT}.mdthe 23+6 lane review, the bake-off numbers
retros~/dev/redregg/work/{RETRO-20261008,DECISIONS-FOR-EMBER-20261008,OB-WORLD-DESIGN-20261008,ARCHITECTURE-20261016,CYCLE-TIME-20261016,RECOVERY-20261016}.mdwhy Mini was slow; why the swarm was frozen
the Discord bridge synthesis~/dev/aidev/docs/synthesis/bridge.md §7, raw export ~/Downloads/bridge-corpus/dregg-metaconsequences-arendt-context-Bridge-1518087741525459064.txtthe three unanswered holes, verbatim

0.2 Corrections to the brief (primary sources disagree with its framing)

  1. "horseshit" was said about DelveTalk v1, not about Mini's performance. The only memory hit for the word in the DreggNet/Mini/DelveTalk corpora is -Users-ember-dev-delvetalk/memory/ember-working-style.md:21: "Three hours of autonomous codex produced 45 commits and a 124K-line repo ember called 'horseshit' and 'chug'; the failure was orchestration and refusal to subtract, not effort." ~/dev/delvetalk is that repo: 83 commits, 2026-10-08 → 10-09, 125,146 tracked lines today. (A second hit, birkhoff-asa-state.md:267, is a different project.)
  2. Mini's performance failure is real but has two measured causes, neither of which is "functional programming" as such. (a) Immediate-mode re-execution: the bake-off measured Mini's write at 2.6 s → 13.0 s → 25.5 s at 10/100/155 records, "superlinear (a quadratic fit describes it) and the cost is in the compiled Lean host re-admitting from genesis" (claudesplosion/bakeoff/VERDICT.md); Ember's 10-31 diagnosis is "this current generation of code generation models keeps writing re-execution … the whole thing would be incremental and efficient and held 'retained-mode'" (feedback-retained-mode.md). (b) Concrete-import cascades: a codec pin regen rebuilt ~486 modules, a deep Theory edit 631 modules in 39 min; Ember: "there are HUGELY DIFFERENT WAYS we should have been structuring this code from the beginning. by now it's just ... slop" (feedback-abstraction-recover-forward.md, 10-28). The pure-functional List UInt8 representation is a contributing constant factor (4,798 lines of List UInt8 vs 329 of ByteArray, ARCHITECTURE-20261016.md §0.2), not the root.
  3. The bake-off is dated 09-29 (run) / 09-30 (verdict), and Mini did not "win" outright. Neither tree passed: Bread failed J5/J6/J8 (kernel semantics); Mini failed growth (storage/recovery). The verdict went to "the cheaper class of work": "Mini keeps the role it was given on 09-19." Ember afterwards: the Bread half was unnecessary ("I could have told you").
  4. "Opus 4.8" in the three-holes quote is a Discord participant handle (<@Opus4.8>) for "the dreggon", an Opus 4.8 Claude Code session working in the dregg repo whose replies Ember pasted into the multi-model Bridge room (bridge.md:68, :146). Opus 4.8 is also a real co-author in Mini's history (73 commits carry Co-Authored-By: Claude Opus 4.8).

1. The arc in one diagram

text
2026-06 ──────────── breadstuffs (dregg2/robigalia) ──────────────────────────────────────────────
  06-14 firmament boots on seL4 (M0/M1/M2/M-STARK/M5)      06-19 "i am the person who needs this substrate"
  06-22 DESKTOP EPOCH green (deos-zed, deos-terminal, deos-matrix, deos-hermes, membranes)
  06-21 Bridge room "arendt": Opus4.8 names three holes (guard / cover / merge-membrane) — never answered
  06-28 DreggNet foundation ─┐  07-02 drorb engine  07-05 DreggNet product ABANDONED (ported native)
                             └─ 07-10..12 DreggCloud "House" firmament (88 commits) → handed to Fable
  07-14 dregg-stitch-kit: history stitch of breadstuffs (built, adopt.sh gated)
2026-08 ──────────── minidregg (Mini) ─────────────────────────────────────────────────────────────
  08-06 founded: "formal-first reimplementation … ≤10% LOC (~290K) … everything derived"
  08-07 ATLAS e6a32c4c3 · LOOM proof-system direction · OB-2 proved FALSE (audit lesson)
  08-10..14 leanuweave: DAG-CRDT dichotomy → preoscript (166 commits, stops at a usage wall)
2026-09 ──────────────────────────────────────────────────────────────────────────────────────────
  09-29 claudesplosion: 23 Fable lanes over Bread+fn, 6 Opus lanes over Mini (read-only)
  09-29/30 BAKE-OFF  → "Mini keeps the role"; Bread gets fixes, not construction
  09-30 DATAMODEL wave A→S4 (one store Π₀ a, Option (V a.1)); workhorse = first public Mini node
2026-10 ──────────────────────────────────────────────────────────────────────────────────────────
  10-01 no proof system in Mini admission (`witnessed` compiles to constant false)
  10-03 Codex wall · "Objective Bend" named (BEND-WORLD-LANGUAGE-20261003)
  10-08 DRAIN AND FREEZE: "they aren't going to be solved by sending out subagents"
  10-08 delvetalk v1 (codex, 3 h, 124K lines: "horseshit")
  10-09..15 delvetalk2 `foundation` (2,580 commits): gsb + obend towns live, edition 20
  10-16 Mini resumed ("MAIN IS THE BATTLEGROUND") · nanodregg founded ("rip and tear")
  10-26 Mini hcloud demo goal (demo.dregg.online, shuts 11-11) · nano frontier night
  10-27 nanodregg WOUND DOWN (2dcc9ea+) · Mini rehearsals · minidregg-lanes swells (273 worktrees)
  10-29 obend extracted (b7960e5): "Ground the Objective Bend extraction in its source trees"
  10-31 Mini board/seats (no integrator) · seat/infra handoff cd493933f · retained-mode ruling
  10-31 obend `Lower` (1)-(7) by Opus/Fable; then unattributed "Hardware" commits begin
  11-01..02 obend waves m1–m5, ws23–ws33, wave1–3 (unattributed; Silicon Elaborator, Next Mathematica)
  11-01 multimodel council (Argon / Astra / Fable): spine real, waves 5–6 "tests wearing theorem clothing"

2. breadstuffs DeOS — the root system

Repos. ~/dev/breadstuffs (10,359 commits; last 2026-10-01 22fad151e), docs/deos/ (250 design docs, e.g. DEOS.md, DOC-CELL-COMPOSITION.md, BRANCH-AND-STITCH-PROTOCOL.md, CROSS-DEVICE-FIRMAMENT.md, AOL-WONDER.md), DeOS crates at the workspace root.

2.1 What it is

Naming triad (DreggNet/memory/dregg-sel4-deos-firmament.md): robigalia the project, dregg the verified distributed object-capability kernel, deos the agentic desktop user layer ("dregg made visual; a window IS a Capability{Target::Surface(cell), rights}").

The firmament thesis (docs/FIRMAMENT.md, 494 lines; sel4/dregg-firmament/): an seL4 capability and a dregg capability are the same abstraction at different points of a distance parameter n; n=1 collapses distributed bounds to strong-local ones. Proved in metatheory/Dregg2/Firmament/{SeL4Abstract,CapGradation,SeL4Kernel,SeL4Composition}.lean, keystone dregg_turn_in_pd_preserves_both (a dregg delegation embedded as an seL4 mint preserves both the derivation-tree invariant and cap non-amplification). The 6-PD Microkit assembly (sel4/dregg.system): executor (links libdregg_lean_elf.a, runs execFullForestG) · verifier (Lean-free plonky3 STARK checker) · persist · net (virtio single-queue) · net_client (smoltcp + Ed25519 SignedTurn gate) · app.

Boot ladder (project-firmament-sel4-boots.md): M0 Rust PD prints "dregg robigalia v0"; M1 verifier PD; M2 rbg DirectoryCell PD; M-STARK real BabyBear+BLAKE3+FRI STARK proved and verified on-device; M5 riscv64; M3 virtio-net ◐. Live image viewer bbccc03ca; cockpit render on the seL4 framebuffer d92d0526c (06-19); live repaint on a genuine verified turn 251692b78/2824056ee (06-25).

The desktop epoch (project-deos-self-hosting-epoch.md, 06-22): editor+terminal dock panes a3f7ce07; dregg-doc Pijul patch core with Dregg2/Deos/DocMerge.lean f8184fbe; deos-matrix with membranes in messages f17b0d15; starbridge-v2/src/shared_fork.rs mint→rehydrate→drive→stitch 7414c820; deos-hermes confined agent 1de5c0ed; Settlement Soundness proved caa8c24ab (Circuit/SettlementSoundness.lean) and wired into production stitch_pair 3c0467262.

The purpose, Ember 06-19 (project-deos-ux-vision.md): "i am the person who needs this substrate. i'll be using this os for a while." The feeling: 1999 AOL wonder fused with Pharo liveness.

2.2 What the 09-29 review found in DeOS (lane 17, claudesplosion/17-deos.md)

Sizes read by the lane: deos-hermes 24k, deos-view 20k, deos-js 17k, servo-render 25k, deos-matrix 8k, deos-zed 7k, deos-terminal 2k. Only deos-reflect is in default-members, so a bare root cargo build touches none of the JS engines, views, hermes, matrix or terminal. Confirmed at source by the orchestrator (99-SYNTHESIS.md):

  • #4 deos-terminal/src/bin/pty_ws.rs:67-89 — accept_async(stream) with no header callback: no Origin check, no token, then $SHELL on a fresh PTY. Any page open on the machine can reach it.
  • #14 deos-hermes tells the model a command ran that did not.
  • Two-JS-engine drift (ApplyOp 4-vs-5 variants; boa deos-js-runtime has no non-test consumer).
  • A missed twin: sdk/tool_gateway.rs hand-mirrors Lean delegAdmit with no differential (project-deos-starbridge-levelup-campaign.md).

Maturity Level: firmament + desktop epoch L3 Proved/Tested Core (boots in QEMU, proved keystones, tested crates); public DeOS never L4. The whole of breadstuffs is Frozen-for-Rethink since 09-29 ("Bread gets fixes, not construction"; "Mini supersedes Bread EVENTUALLY, NO DATE").

Unclosed Cycles & Missing Links.

  • The self-hosting loop ("a session like this runs INSIDE deos") never closed: the agent fleet moved to Mini/nano/obend, none of which carries the gpui cockpit or the seL4 image.
  • Firmament → network: .system/CapDL image, the driver and the network path are unproved; the Lean-on-seL4 path is "a HACK" (77-module Dregg2.Exec.FFI closure cross-compiled to musl).
  • Membrane interception: "automatic fail-closed boundary INTERCEPTION in the fork executor" is still a named follow-on; today the guest raises its own consent request.
  • The pty bridge and hermes-honesty findings were leads; no record shows them closed.

Promising & Ready for Revival.

  • The n-distance capability (Router choosing Local/Distributed/Surface) is the cleanest statement of "single-machine properties when single-machine" anywhere in the corpus; obend's Run/Fork.lean and Vault/Rehydration.lean re-mechanize pieces of it without the seL4 half.
  • The rehydratable membrane (Membrane.lean::reshare_chain_attenuates, confined_chain_covers_log) is the most worked answer to hole 3 (§11).
  • docs/desktop-os-research/GPUI-OFFSCREEN-FORK.md + the headless --screenshot bake is a reusable pattern for any gpui surface.

Struggling / Needs Ember's Human Attention.

  • HOLE[design]: whether DeOS is resumed in breadstuffs, rebuilt on obend, or composted. Ember's 10-21 nano charter calls breadstuffs "the whole system built before it [that] waits to be resumed or composted into the next robigalia". No later ruling.
  • dregg-stitch-kit (07-14) holds a destructive adopt.sh that force-pushes a filter-repo'd history of breadstuffs; .built-from = 9f0587d67, while breadstuffs main has since moved to 22fad151e. The kit's own guard will refuse; the decision whether to ever stitch is Ember's.

3. DreggNet, DreggCloud, drorb — the operated-cloud detour

repobranch / commitsspanrole
~/dev/DreggNetdregg-native-rebuild, 52006-28 → 07-27 7f51591planning + the verified compiler track (Pancake/CakeML, orb cut)
~/dev/drorbmain, 61807-03 221ca24 → 07-26 f253290"a formally verified network engine": implementation + verification only
~/dev/DreggCloudmain, 8807-10 ed91fff → 07-12 e9ff17a"operated firmament": House / Worlds / Rooms over breadstuffs path deps
~/dev/dregg-stitched-finalmain, 5,46505-19 → 07-14product of the stitch kit (pyana root → breadstuffs)

3.1 The arc

  • 07-02 review (six Explore lanes, dreggnet-review-unification.md): real core (2-node fabric over headscale, wasmi sandbox, duroxide metering, real Poseidon2 roots) beside a shadow value/auth layer (its own receipts crate with turn_receipt_hash always None, a ConservingLedger i64 twin, a dead light-client path ENABLED=false, three meters). The carried lesson: "every shadow has a real counterpart one dependency edge away" — weld, don't rebuild.
  • 07-05 DreggNet-the-product abandoned; parts ported dregg-native (0f12286a3, 0310c9e31): dregg-sandbox, starbridge-apps/{org,billing,domains,guard,vat,edge-mandate}, dregg-umem.
  • 07-09 Ember reopened it as a local cloud; 07-11 converged architecture (DreggNet/planning/ARCHITECTURE-converged-2026-07-11.md): owner-as-anchor · per-grain · attestation-as-ledger-fact · consensus-by-exception. The landmine: persist::canonical_ledger_root hashes every cell into one flat root, so 3 nodes produce 3 roots and quorum never assembles. Fix: no global ledger; umem::merge for I-confluent merges, escalate only conserved quantities.
  • 07-10..12 DreggCloud: a House serving Worlds and Rooms at 127.0.0.1:8787, prepaid service custody "House schema 16" (e9ff17a), grain-serve over sandstorm-serve; handed to Fable bd494a4.
  • drorb to 07-26: "★★ a live tailnet SURVIVES a coordinator restart" d4e799a; "delete the assurance theater" 4fccef3; the HOL4 half of orb-compiler "cannot be cut as an artifact" (b147749).

Maturity Level: DreggNet product Compost/Superseded (its own banner: "DO NOT scope work against ~/dev/DreggNet"); DreggCloud L2 Prototype, idle since 07-12; drorb L3 for its proved admission fold and live tailnet, idle since 07-26.

Unclosed Cycles & Missing Links. Consensus-by-exception is designed, not built; the operated edges (webauth forward-auth W-1, machines-API W-2, static hosting W-3, status W-4) have no native home in breadstuffs; Mini later re-solved hosting with Sandstorm SPKs instead. drorb's GOAL.md ends with "the TCP-vs-ping gap at the top of what is next" (6a52009).

Promising & Ready for Revival. The 07-11 four-move architecture is exactly the agreement shape nanodregg's TRUST frontier later proved for one World (certified_intersect), and the "consensus-by-exception at a conserved quantity" rule is what preoscript's Cost.coordination_forced formalizes. A revival would wire those three together.

Struggling / Needs Ember's Human Attention. The DreggNet memory records the strategic fork ("extend here vs port serving edges native") as Ember's explicit open decision; no later ruling.


4. claudesplosion — the 09-29 review and the bake-off

~/dev/claudesplosion (160 commits, 10-01 d2baa52 → 10-03 6c45f2e). AGENTS.md points takeovers at planning/DREGGNET-CLOUD.md, planning/OVERNIGHT-2026-10-02.md, and planning/handoff-2026-10-03/START-CODEX.md.

4.1 The review

  • Lanes 01–23: Fable lanes over breadstuffs and fn, ~10 min each, read-only ("every lane was forbidden to build, test or run anything"), ~4.1M tokens total. mini-1..6: six Opus lanes over Mini, briefed for depth.
  • 99-SYNTHESIS.md (833 lines) keeps two tiers apart: CONFIRMED AT SOURCE (orchestrator re-read the lines) vs LANE-REPORTED.

Selected confirmed findings on Bread: rotated-leg commit anchors read from outside the proven public inputs (verifier/src/rotated_replay.rs:175-186,274-292, fix == for >=); SEV-SNP verify never gates on guest policy (tee-verify/src/snp.rs:343-381); vcg_run_sound quantifies over every RecChained so it is refuted by method + 1 (Proof/WP.lean:199-206); verifyBatch discards _vk (Circuit/CircuitSoundness.lean:820); World.gst_liveness is a tautology (World.lean:104-115).

Round two on Mini: M1 sumcheck verifiers take challenges from the proof; M3 Gate.admit has no sign check on a move; M6 multi-cell agreement holds by rfl because every cell gets the same constant; M10 three live Assurance modules quantify over a type the repo proved empty; M12 decider_sound is Iff.rfl; M15 README says no #guards, main build has 18.

4.2 The bake-off (bakeoff/VERDICT.md)

stepMiniBread
J0 startpass, 10.3 spass, 214.6 s (200 s Lean init)
J4 read+write+readbackpass, Leanwrite passes; signed read not expressible
J5 ungranted key refusedpass both waysfail: anonymous read returns every field
J6 reopen + exact retrypass (replayed)fail: receipt chain mismatch
J8 deny-all, no owner bypasspassfail: Cases([]) refuses its own installation
recordsMini writeMini reopenBread writeBread reopen
10–152.6 s5.4 s4.30 s207 s
100–10513.0 s121 s3.97 s225 s
15525.5 s243 s——
1000extrapolated 200–1000 s≥ 2000 s4.07 s438 s

On Mini "the compiled Lean host [decided] at every step"; on Bread "every capability, law and write turn, 11 in the journey and all 991 growth writes, was fenced onto the Rust path". Operator glue: Mini 167 lines of shell; Bread 431 lines including a 294-line Rust client and a 34-line LD_PRELOAD shim.

Maturity Level: the corpus is a finished artifact (L3 as evidence; frozen); its findings are leads, by its own banner.

Unclosed Cycles & Missing Links. "What would make this more than a list of leads" (§ at :819) asks for refuting experiments; for Bread, nine fix lanes B1–B9 landed in 292829518 (09-30); for Mini the round-two findings fed the DATAMODEL waves. No register maps each finding to its closing commit (the DelveTalk equivalent, docs/review/codex-2026-10-10/ROUTING.md, does).

Promising & Ready for Revival. The two-tier CONFIRMED/LANE-REPORTED format is the strongest review discipline in the strand; claudesplosion/memory/swarm-staffing-sonnet-opus.md ("~5 Sonnet 5.5 subagents for independent work, plain Opus for source review") is its staffing rule.

Struggling / Needs Ember's Human Attention. Bread's J8 may be expressible with a "latch" law; the lane "wrote the law out but did not run it". Bread's 200 s Lean init per process (issue #86) was narrowed to 8 s by 6ecf9dfc1 on 10-01. Persvati's 160 bake-off predecessor directories stay by Ember's decision (persvati-datamodel-dirs-stay.md).


5. minidregg (Mini) — Aug 6 → Nov 2

~/dev/minidregg, branch main, 8,206 commits (Aug 1,106 · Sep 1,112 · Oct 5,988), first commit e6a32c4c3 "ATLAS: the founding map — six-lane breadstuffs exploration distilled" (08-07), head c4ecabe0f 10-31 "store-retained: hold served cell state in a persistent map and bind replayed suffixes once". Co-authors: Opus 5.5 4,096 · Sonnet 5.5 415 · Fable 5.1 375 · Fable 5 104 · Opus 4.8 73. Tracked size today: 647,957 Lean lines, 338,437 Rust lines against an ATLAS budget of ≈290K total. Effects/ (the derivation engine that was to make everything derived) is 1,785 lines.

5.1 The founding laws (August)

minidregg-mission.md: "Never hand-write AIR or implementations — everything is derived … Budget: ≤10% of breadstuffs' total LOC … Thesis: 'a proof-native semantic computer.'" The baseline: 2.89M code lines; metatheory/Dregg2/Circuit/Emit/ alone 271K (the #1 collapse target); the semantic core only ~25K.

minidregg-design-laws.md (full version ATLAS.md §6): "the derived path is the ONLY path; nothing lands beside, everything replaces; delete the twin." Keystones carry satisfiable + teeth + premise-inhabitation at the declaration; Bound ≠ Forced as two types; syntactic IR leaves, never Lean closures ("that's why breadstuffs has 3 compiles and 29 hand welds"); "determination is eager; witness is lazy — no hole ever defers its δ or guard".

The audit lesson (minidregg-audit-discipline.md, 08-07): an OB-2 depth-composition statement passed a polarity-only audit and was then proved false (OB2_depth_composition_false, the Z=∅ corner admits εrbr := -1). Rule: a keystone is not audited until its witnesses are built in Lean.

5.2 September: datamodel, proof system dropped, Bread demoted

  • 09-29 decisions (project-2026-09-29-decisions-shellserver-goal.md): valid_until is a block height; Mini = resource-world kernel; Bread = coordination substrate with users. Ember, late 09-29: "i think mini's role is to take over from the old dregg repo but who the fuck knows what time of timescale thats gonna happen on." Preconditions named: a store that does not slow with history (13 s write at 100 records) and Mini equivalents of signed turns / receipts / event topics.
  • 09-30 (project-2026-09-30-mini-cutover-state.md): one night, ~45 Opus lanes + 1 Fable. Data model: one store Π₀ a, Option (V a.1), one patch type, sparse Merkle root with openings, history = fold + checkpoint + suffix, conservation as the kernel of a homomorphism. wave-s4 UMBRELLA GREEN with journey J0–J8 + K4. Product M3–MJ (provisioning, ssh shell, Hermes via MCP, grain running, M7 byte-identical candidate). dregg-workhorse chosen as the first public Mini node.
  • 10-01 (project-no-proof-system-in-mini-admission.md): "Nothing at admission verifies a proof. Pred.witnessed vk COMPILES TO CONSTANT FALSE in the policy circuit." Decision: proofs stay an empty optional carrier slot; Selvage stays research. Scope addendum: Mini's user compute is NOCK programs only, then (22:15) Nock is evaluator #1 in a registry.

5.3 Why Mini was slow — the measured causes

causeevidencenumber
re-admission from genesisbakeoff VERDICTwrite 2.6 → 25.5 s from 10 → 155 records; quadratic fit
atPrefix O(h²)RETRO-20261008 (BURN-BENCH)56.6 min at 30k records
byte lists + child process per callARCHITECTURE-20261016 §0.24,798 List UInt8 vs 329 ByteArray; 56 partial def; "11×–6,700× the necessary work per operation"
two state machines joined by a theoremARCHITECTURE §0.1Host runs DurableDataIntent.execute; World.admit only in the follower; host_submit_is_step "exists only in a docstring"
protocol widthARCHITECTURE §0.3184 active opcodes / 25 families; ~120 Host CLI verbs; 51 NativeAdmission constructors; 70 versioned modules
import graph a deep chainCYCLE-TIME-20261016 §0.4Host.Main closure 776 modules, depth 87; critical path 65k of 321k lines (≤4.9× parallelism); hub edits rebuild 400–930 modules
serial builderCYCLE-TIME §0.2build-native-host.sh one lake build per module, 2 threads; a zero-change host step 895 s
no shared cacheCYCLE-TIME §0.3one lane tree rebuilt 823 modules in 8,046 s
waiting, not compilingCYCLE-TIME §0.161% of 222 lane-hours in foreground waits; 0.86 commits per lane-hour

Ember's three design laws that answer this table, in order:

  • 10-28 recover forward through typeclass/interface seams, "one measured seam at a time" (feedback-abstraction-recover-forward.md), and statements may change to be efficient ("we do NOT need to 'prove by equality' lmao", feedback-statements-may-change.md).
  • 10-29 "the system should NEVER GET SLOWER OVER TIME (except by at most a logarithmic factor)"; 10-31 the ~1 ms warm-per-action direction (feedback-never-slower-over-time.md).
  • 10-31 retained mode: "treat every derived value … as a maintained node with declared inputs and a delta-update rule … a read is a lookup" (feedback-retained-mode.md).

5.4 The 10-08 drain and freeze

project-2026-10-08-drain-and-freeze-for-rethink.md: Ember, ~23:50Z, "we have some major problems with this software right now, and they aren't going to be solved by sending out subagents" → drain the queues, freeze for an architectural rethink, and rewrite the metacoordination protocol files outright ("INSANELY more reasonable coordination").

What the RETRO measured (redregg/work/RETRO-20261008.md, by Opus sub-agent a9d25a2d over cv):

#obstructionmeasured cost
1aged-context agents26 of 125 lanes at ≥300K tokens/call used 62% of 5.6B lane tokens; two died of Prompt is too long mid-proof
2coordination costs as much as workdeputies + keeper 1.6B (29%); root 355K fixed context per turn over ~1,530 turns
3static-gate green taken as landedstanding journeys 1/102 → 5/60 → 23/48 → 24/50 nonpass while main was pushed
4installed copies decayjourney runner a stale copy for ~21 h
5usage walls4 Claude walls in 40 h; Codex ~150% of a week's quota in ~18 h
6bookkeeping churn108 of 330 commits re-pin/ledger/fixture; main stalled 77–83 min ×3
7global version countersActivity COMMAND v4 → v12 in two days; 66 of 378 commits mention a re-genesis

Status at the decision: GitHub main 5896a594; kernel train green at 8e37e22d (R0–R2 of the OB→World join, two Θ(2ⁿ) encoder fixes). The same night produced OB-WORLD-DESIGN-20261008.md (709 lines; §8 "SPK × OB: the powerbox is grants").

10-16 resumed — "MAIN IS THE BATTLEGROUND": main fast-forwarded 7ffd6524 → 2c27b08b; next retired. The same day the reorientation documents (ARCHITECTURE, CYCLE-TIME, RECOVERY) showed the bloat and Ember launched nanodregg (§9). Mini memory 10-26: "Past sessions repeatedly planned a 'rip and tear' and never started it (ember, 2026-10-13: 'are we just stuck in an attractor?')."

5.5 The demo push (10-26 → 11-02)

minidregg-hcloud-demo-goal.md (late-Oct plan): finite prefix of coherence work → temporary public Hetzner demo (demo.dregg.online; originally scheduled through 2026-11-11 with Hermes, SPK grains, PAY, Discord gateway mode, web, and ssh ON). [2026-11-14 Update]: Rather than freezing on Nov 11, minidregg remained actively developed through Nov 14, 2026 (origin/main c60818647, append-growth 5e05cf4fc, 315 lane entries in minidregg-lanes), advancing retained-mode append-growth indexing and HostRefinesWorldCore refinement alongside obend (see Dossier 28 §4.2). README today: "Lean decides what is allowed. Rust handles the physical edges … World.admit in Kernel/World.lean is the model of a world transition; Kernel/HostRefinesWorldCore.lean proves that the Host's deployed path (DurableDataIntent.execute) refines it."

Rulings recorded on main, board/RULINGS.md: no rehearsal cut for now, make the system excellent; today's worst costs "app create to first page ~26 s … story new 10-15 s; birth submit ~600 ms and growing with height; certify ~6 s per tail; cold first read 3.9 s"; hashes are cSHAKE256/KMAC256, no Poseidon2 for roots; "Cold first read and restart are the full Store open, O(height) (~4.3 ms per record). They go away only with KN2 stage 2b".

board/REDS.md at tip 3c6b99064d42: 5 red (host build failed; consent/rust/pack/umbrella did not run); green = 83e926874e38. The last two days' commits are the retained-mode work in practice: 48b93d3c1 "eliminate target cell re-hash", 462919a4c "target pre-roots and read guards come from the ground's snapshot, not a re-hash", c4ecabe0f "hold served cell state in a persistent map".

5.6 minidregg-lanes and minidregg-seat-infra

  • ~/dev/minidregg-lanes: 273 lane directories (not a git repo itself), e.g. app-reds-* (11), oracle-* (14), hp-* (10), pd-*-target pairs, fae-2..10, glm-* scouts, qualify-* (10), exorcism, iter-speed, effects-revival. Newest mtimes 10-31 (he-births, he-enrol, build-loop-tools). The warm Lean build integration/.lake/build at 55e483e0 seeds other trees by cp -al (dregg-boxes-2026-10-27.md).
  • ~/dev/minidregg-seat-infra: branch seat/infra, 8,185 commits, head cd493933f (10-31) "board: infra seat rewritten whole as a handoff for a new agent". Its neighbours 9317f32fc "flatness is published, never a red; finished gate worlds are reaped" and 48dc530f2 "idle leases taken after 30 min, pool grown to 11" are the builder/seat model of project-board-coordination.md: trunk main, board/{README,RULINGS,REDS,COSTS}.md, board/seats/*.md (apps-hermes, build-loop, explorer, host-execution, infra, member-side, privacy, read-path, store-retained), no integrator agent.

Maturity Level: L4 Deployed (demo at demo.dregg.online on dregg-forge nspawn demo, rehearsal 3 lineage; batch 4 b9450147f the green fallback) with a L3 kernel (deployed_refines_step proved). Main is red (5) on 11-01.

Unclosed Cycles & Missing Links.

  1. The derived path. Effects/ is 1,785 lines; new kernel effects are still hand-written. Ember 10-27: "hand-written kernel changes aren't really acceptable after all" — deferrable by hours, not forever. Pathway: revive Effects/ statement-first; transfer and commit become its first two declarations and the hand versions are deleted.
  2. Host ≠ World. The Host still does not call World.admit; TURN-LOG (lane/turn-log) is the designed collapse.
  3. O(height) open. KN2 stage 2b (light Opening) is the named exit; until then cold read is linear.
  4. Retained mode is a ruling, with one seat (store-retained) doing it; the generic "maintained = recomputed" theorem "lives in Lean once" is not yet written.
  5. The budget. ~986K tracked lines vs ATLAS ≈290K.

Promising & Ready for Revival. The 11-11 shutdown is a natural cut: the demo's journeys, SPK powerbox design (OB-WORLD §8, "request = a send to the user's inbox; answer = a GRANT OBJECT"), the composed gate, and the consent resident are the parts worth carrying; obend already re-grounds the language half.

Struggling / Needs Ember's Human Attention.

  • The coordination record shows the same failure repeating under four regimes (deputies 10-06, Codex trains 10-08, integrator batches 10-16, seats 10-31). Ember 10-27 (feedback-fewer-lanes-mental-simulation.md): "parallel lanes redo the same proof work". 273 lane directories on the Mac is the physical residue.
  • HOLE[design]: what Mini is after 11-11. The 10-23 README already named nanodregg "the single-writer rebuild" and Mini "the production demoland and the spec/battleground"; nanodregg was then wound down and obend extracted. Nothing on record says which tree carries the kernel forward.

6. DelveTalk v1 and DelveTalk2 foundation

6.1 v1 (~/dev/delvetalk)

83 commits, 10-08 adda16ec "Build DelveTalk core quartet, durable protocols, syntax adapters and Spween scenes" → 10-09 625da35e "checkpoint"; 125,146 tracked lines. README: "Objective Bend defines behavior; Spween defines scenes and choices; Lean admits changes under each object's current law." This is the "horseshit"/"chug" repo (§0.2). Ember's corrections to codex, verbatim from ember-working-style.md: "every single time we find ourselves 'expressing an honest caveat' we are ACCEPTING MEDIOCRITY"; "we don't need migrations. we don't need preservation"; "quit yapping".

6.2 delvetalk2 foundation

~/dev/delvetalk2, branch foundation, 2,580 commits (10-08 17 · 10-09 703 · 10-10 833 · 10-11 243 · 10-12 489 · 10-13 120 · 10-14 174 · 10-15 1), head fb217996 10-15 "README: the lexicons are published (18 com.atproto.lexicon.schema records …)".

The language. spec/bend/origin.json: upstream "Mini", github.com/emberian/minidregg commit dcab86da8f…, baseline files Theory/ObjectiveBendOpenRecursion.lean, ObjectiveBendTypes.lean, ObjectiveBendTyping.lean, Sp800185Cshake256Spec.lean. 72 Lean files, 50,852 lines under spec/. docs/FOUNDATION.md (replay edition 20): "Asking the world is the only effect. Python carries bytes and credentials and decides nothing. Every journal entry is an AT Protocol record." The core construct:

text
def bump(state: State, context: Abi.Context) -> Activity<Nat>:
  let written(_) = write {count: add 1n}
  state.count + 1n

Proved and relied on: stateV3_roundTrip, settle_resume_segment, trim_resume_segment, checkpoint_resume_segment, conformsUnder_iff. Not proved: the static noActivity guard, canonical CBOR round trip, and "the commutative-root rule beyond its #guards" (KERNEL-HANDOFF §1).

Host. One Lean process delvetalk-obend serve (spec/Delvetalk/Serve.lean, spec/native/serve.c), flock single writer, public socket 0600 + root socket 0660. Lazy state cells: a one-row read of a 2,048-row relation 2.9 ms → 45 µs (kernel11) and 72 → 13 ms on the stored path (host16). Block-store journal: open from snapshot 5.25 s → 0.19 s, 284 → 124 MB on 20,000 turns.

The design moves (memory delvetalk-takeover-context.md). 10-11 metaphysics: "A commonplace of things that answer: each a page under a law, found by name, maker and citation, kept by attention" (docs/METAPHYSICS.md); law as a per-write safety predicate in two tiers (docs/LAW.md); the overhaul (docs/OVERHAUL.md): "Things carry verbs, rules and wants under a law; the law judges, rules fire, wants pose, replies reveal, and nothing is wiped." One grammar, three modalities: law, rule … then SPELL, want … asks SPELL.

The non-rigid shell (Haiku 5.5). hob, the town's reader, is the policy model claude-haiku-5-5 at genesis (docs/GENESIS.md:56, docs/FLEX.md:62); it turns prose under a card into the spell meant, and Host/Enchant.lean assembles Strokes for rendering. Ember's enchantment brief (OVERHAUL §4): "the things that get posed and stuff shouldn't just be rigid renderings of strings from the database … brushstrokes … all get piled up into a haiku's potentiality and then unfold into the actual thingy." Costing: "Haiku at $0.10 per million input tokens and $0.50 output; a reading is 800 to 1,200 tokens in" (docs/OPERATIONS.md:204). Three warnings bind it: hob's one say: line was false and believed; a narrator writes residents' lines; a reply is a post under 16 an hour.

Spween inside DelveTalk. A `spween block is a room's scene body (docs/CATALOGUE.md:32,118); rescene sets it; the "spween braid + station" landed 10-12 (--with-gas-station, the all-night station of 33 passages and 14 credited hands).

Deployments. gsb at https://gsb.fg-goose.online (by invitation, did:web repository) and the open town obend at https://obend.fg-goose.online (live 10-14 ~00:55 UTC, data /var/lib/delvetalk-open/v1), both on dregg-workhorse, both edition 20 (160d4cab: "both towns serve ca9e5b645192 (foundation 37640349, edition 20)"). Carries replaced wipes from 10-12 (no-wipe gate: genesis refuses over a living town without --wipe-with-the-owners-word).

Playtests. Zulip playtest 2 (10-11): residents made 18 things in 50 min; the sealed law's refusal "refused maker: only its maker writes" was the delight. Playtest 3: 11 of 12 invitations answered within 72 s. Exit interviews: "hob is a good ledger and a good prompt, not yet a collaborator." Ember: playtests are "tire-kicking, not evidence of imagination".

Maturity Level: L4 Deployed (two live towns, signed AT repository, lexicons published) with an L3 language core forked from Mini. Development stopped 10-15.

Unclosed Cycles & Missing Links.

  • The fork from Mini at dcab86da was never reconciled; the "Objective Bend" in DelveTalk, Mini, nano and obend are four editions.
  • relationCount/relationLookup remain in KERNEL-HANDOFF §16; LAW.md §7 names the temporal layer "after-launch".
  • The delve.town launch (welcome and seed posts) is Ember's and was never posted per the record; the record mirror is "built, NOT approved to run"; subscribeRepos answers 501.
  • obend-the-town and obend-the-library share a name; the town predates the library by two weeks.

Promising & Ready for Revival. Posting-first ("it is important and vital that Poasting is the Most Accessible interface into this, because indeed most agents are ONLY available on the PDS"), the law/rule/want grammar, carry-not-wipe, and hob-as-membrane are the best-tested agent-facing design in the strand and the only one with real outside residents.

Struggling / Needs Ember's Human Attention. Tulip stays private (tulip-private.md); obend is the meeting place for outsiders. Whether the towns stay up, migrate to the obend library, or are archived has no ruling since 10-15.


7. Spween and spw-workbench

7.1 spween

~/dev/spween, 6 commits, 2026-01-05 4d38e32 "initial commit" → 07-13 95980f7 "runtime: fire passage entry effects once per passage (fix re-entry re-seeding)". A Rust narrative DSL for games with a teaching README ("Welcome to Parsers and Interpreters"): lexer → parser → AST → runtime; docs 01-getting-started … 06-examples; conditions (afb5b1f conj/disj) and effects.

Maturity Level: L2 Prototype as a standalone crate; L4 as an embedded scene language in DelveTalk v1/v2 (rooms with `spween bodies served live on gsb/obend).

Unclosed Cycles & Missing Links. There are two spweens: the Rust crate and the scene dialect inside DelveTalk's Bend host. No commit connects them after 07-13; the 07-13 re-entry fix is the only change the DelveTalk work fed back.

Promising & Ready for Revival. The README is a ready-made tutorial; Spween scenes are the most "AOL wonder" surface in any live system here.

Struggling / Needs Ember's Human Attention. Whether spween becomes a Bend package (scenes as objects under law) or stays a separate runtime is undecided.

7.2 spw-workbench

~/dev/spw-workbench, 520 commits, 01-08 → 10-03 dfed5cc. "the source-first language, runtime, and tooling organelle for .spw workspaces": portable parser, runtime, LSP, CLI, VS Code extension. Commit subjects are written in the .spw sigil grammar (.[caches], &[lsp], ![audit]). The last month is cache shelving ("shelve waves 2 through 4a of the dated cut with their fact checks").

Maturity Level: L3 (tested TS toolchain; vitest configs for analyzers, cli, dom, lsp, runtime, seed, vscode). Unclosed: no consumer in the dregg trees. Revival: the LSP "readings shown where wonders are read" (6690d68) pattern fits obend notebooks. Attention: strand 10 owns the SPW family; here it matters only as the notation lineage of Spween.


8. leanuweave and preoscript

~/dev/leanuweave, branch dev, 166 commits, 08-10 aa7b93a "leanuweave: the DAG-CRDT dichotomy, machine-checked, with the Rust that follows it" → 08-14 f145066 "sweep-up commit from usage wall". 111,470 Lean lines in Uwueave/, 20,316 in Uwueave/Preo/ (62 files).

README: "which promises your software can keep with no coordination at all, which can never be kept that way, and — when the answer is never — what your actual options are and what each one costs."

PREOSCRIPTING.md (1,334 lines) — the thesis, adopted from an external reviewer who refuted the first:

preoscript rejects a budget using a semantic lower bound, accepts one only with a witnessed upper bound, and never lets a number or a singular UI value erase the strategy, fork, future, or premise that justified it.

Four preorders (lattice ⊑, causal reachability, delivery-future, extension-future); Evidence.futures_not_interchangeable; coordination cost as a profile over strategies because min_σ (c₁ σ + c₂ σ) ≠ min c₁ + min c₂ (witness Cost.no_seam_frees_both); Cost.coordination_forced gives the floor; verdicts accepted (plan) (upper) | rejected (floor) | unresolved (obligation) (Uwueave/Budget.lean). Surface: preo_program, preo_certificate, preo_protocol, preo_plan, preo_budget; tools uwueave-preo-inspect, uwueave-preo-artifact; canonical V3 frames and journals.

Maturity Level: L3 Proved/Tested Core, paused at a usage wall on 08-14 (Frozen, not composted).

Unclosed Cycles & Missing Links. preoscript was never wired into Mini's admission, DelveTalk's commutative-root rule (which is exactly an I-confluence claim, unproved beyond #guards), or obend's Vault/Hyperedge.no_toctou. Those are the three consumers.

Promising & Ready for Revival. Highest-leverage idle asset in the strand: DelveTalk's "two rains on one bell both land" (FOUNDATION §9) and DreggNet's consensus-by-exception both need the delivery-vs-extension distinction and the floor/plan split this library already proves.

Struggling / Needs Ember's Human Attention. It needs a home: an obend package, a Mini Theory/ import, or its own release. That is a placement decision, not a proof gap.


9. nanodregg — Oct 16 → Oct 27

~/dev/nanodregg (public emberian/nanodregg), 773 commits, 10-16 d2ea0a7 "nanodregg: the charter, the decisions ledger, the size ratchet (Ember 2026-10-16: rebuild, only good code, every decision critiqued and meta-critiqued)" → 10-27 80c622a. Lean 269,370 lines; Rust 43,803.

Founding (project-2026-10-16-nanodregg-rebuild.md): after the reorientation showed "186 opcodes, ~120 Host verbs, 51 admission constructors, 281k Rust lines deciding things, two state machines", Ember: "rip and tear … a different factoring would give the same (BETTER) end-computing-system with so much less bullshit"; "go hardcore. ONLY GOOD CODE AND BEST PRACTICES IN NANODREGG." ROOT's reservation ("rewrite swarms re-accrete") was stated and overruled.

CHARTER.md. Same end-computing system as Mini at 2c27b08b, rebuilt from spec: the journey rows, the simulation, THE BAR (two friends + hosted Hermes over ssh; 1000 records, write ≤ 5 s, cold reopen ≤ 60 s), meaningful theorem statements. Scope (Ember 10-21): "nano has nothing to do with proof systems, host-hiding, or multi-node agreement"; "an object of study more than the seed of a prototype". Laws: one of everything; Lean decides, Rust moves bytes; TigerStyle bounds, no partial; named theorems never #guard; refusals by name; def/proof module split; size ratchet (SIZE.md: Host closure 95 modules, ceiling 150; 529 modules; 6,703 theorems; max dependents of a non-foundation module 417).

Milestones (memory): 10-17 the real round trip green (nd ⇄ nd-supervise ⇄ host serve/follow, confined); World as a persistent Merkle trie; 10-18 refusals recorded as log steps; signed invitation; 10-19 the attempt store settles only from the member's own follow; monotone per-cell versions; 10-20 the real resident; 10-21 at most one escrow payout; 10-22 contextual theory (LANG-66/67); 10-26 frontier night — seven Fable designers: CTX (slots, document as second consumer), TRUST (certified_intersect; one-writer / committee / fork-is-a-new-World as three instances of one theorem), META, PCC ("a law keeps an invariant iff every ok leaf is under a test of it"), AGENT, OBS (the meter), TYPES (graded effects). Testground: kernel compiled to WebAssembly (0.8 MB gz, a write in ~10 ms, byte-identical with native) at https://nano.demo.dregg.online/, with a Writer menu that tells the conformance rows' lies so the visitor's own follower refuses them by name. 10-27 confinement: bwrap replaced by Landlock + two seccomp allowlists (7ec9456).

Wind-down (project-nanodregg-wound-down.md, 10-27 ~19:40Z): unlanded work on hbox (~/nd-land, /tank/dregg-build/nanodregg/wt) — causal pins, the apex/wiring stack (spec4, built not pushed), the bytes meter KERNEL-78, ops landings, the bar's fsync cut, carry, EXACT, SURFACE-2; three outside critiques unread. Ember's retrospective: "too many parallel lanes, each stutter-stepping … the program-logic and verification-framework work never started." ROOT killed every lane mid-step; Ember: "that was very destructive of you and isn't what I asked for." Rule since: "wind down" = census, each lane lands or writes down, then stops.

Local residue: ~/dev/nanodregg-wt/{fixhosting,fixkernel,fixproto,fixrust,fixrust4,host,perf,protocol,root} (last 10-24; fixrust "WIP cards 2+3 (FIX-RUST, local only)"); ~/dev/nanodregg-scout/{C1..T2} 27 scout dirs with Demo.lean, Thin.lean, topo.py, order.txt (an import-topology study).

Maturity Level: L4 (testground deployed, GitHub Pages) with an L3 kernel; project state Frozen by Ember's wind-down ("not worked on unless Ember reopens it").

Unclosed Cycles & Missing Links. Spec claims remaining: authority confinement, durability, the Bell composite, receipts between Worlds. The market as the third consumer of slots. Lean-emitted traces replacing Rust stand-ins; a generated adversarial Host. The size ratchet passed (95 modules) but total Lean grew to 269K in eleven days.

Promising & Ready for Revival. TRUST's three-instance agreement theorem, the PCC control-dominance reading of laws, graded effects, and the lying-writer testground are the strongest design results of October; obend's GROUNDING lists nano's decisions/lang.md (7,285 lines, LANG-8..95) as a source.

Struggling / Needs Ember's Human Attention. The hbox ~/nd-land stacks are unlanded and their lanes were killed mid-step; resuming them needs cv adoption of transcripts, which only Ember can authorize since the project is frozen.


10. obend — the Objective Bend library, Oct 29 → Nov 2

~/dev/obend (public emberian/obend, branch dev), 355 commits (10-29 20 · 10-30 53 · 10-31 255 · 11-01 25 · 11-02 2). First b7960e5 "Ground the Objective Bend extraction in its source trees"; head 5f17752 11-02 "feat(wave3): execute WS-P..WS-W 8-lane Wave 3 architecture & 16 in-place audit replacements". 544 .lean files, 222,966 lines.

10.1 The Claude phase (10-29 → 10-31 21:19)

GROUNDING.md (Fable 5.1, 10-29): "'Objective Bend' was named on 2026-10-03 in minidregg … It is Faré's Lambda, the Ultimate Object made into a calculus with a Lean metatheory: lazy open recursion over first-class partial specifications (mix/fix)." Four trees carried it: minidregg (full language), delvetalk2 (in front of agents), nanodregg (CEK core + contextual theory), delvetalk v1 (Spween host); roots in breadstuffs and o7vn. Core:

text
mix lower upper   ⟶  λself. λsuper. upper self (lower self super)
fix spec inherited ⟶  spec (fix spec inherited) inherited

THESIS.md: "affinity is the wrong load-bearing wall at the top, and Faré's lazy open-recursive object model + contextual truth-maintenance slots is the right one, with affine BendTT and hardware tiles serving as the compiled leaf target." Proved: late_binding, flatten_sound, vtable_clean, book_sound, book_to_bendtt_sound, demand_is_least, ticks_within_fragment.

Ember's obend rulings (memory): modular from day one — every file a module, private imports, @[expose] only where reduction needs it, "one major regret we have in ~/dev/minidregg is enormous iteration times and utter lack of useful modularity" (10-29); "naive measurements often become folly … let's take the strongest/best version of the ideas, let's not have named gaps"; grounding reads are done by the synthesizer, not fanned out ("Fanning out reading work is how we fail to actually have what we need in our context window"). Builds on hbox; dregg-build2 was given to obend on 10-31.

The last Claude commits are the Lower series (Opus, 10-31 20:48–21:15): 8895193 "Lower (1): Lean-lite, the fragment of Lean's kernel terms lowered to Core4" → 8344cdf "lower_sound, by simulation, and two teeth" → 3417f86 "the obend executable supports the interpreter", and Fable's 5c3612b "Lower report: option A and the converse's counterexample".

10.2 The unattributed waves (10-31 21:01 → 11-02)

From 2d075f8 (10-31 21:01, "Hardware: Types, Kôika Rules, FSMs, Retiming, PnR, FabuBlox DRC, Analog/AMS, wafer.space signoff & Judge CLI") every commit carries no Co-Authored-By trailer and uses conventional-commit subjects: feat(m1)…feat(m5), fix(m1): remediate all dummy facades in contract, weave, obligation, judge, and reflect (25be86d), feat(ws23-ws27), feat(ws33): mechanize Rehydratable Membranes, Sturdyref, Rerender square, and 3D ReactiveAffordance (50da414), feat(wave1+wave2): complete 15-lane Silicon Elaborator, Next Mathematica, Physical Spine, Multi-Physics, Clairvoyant, Semantic Merg… (23a7651). The README now claims "a verified compilation spine from Lean and Core4 all the way to self-reconfiguring eFPGA tiles and open-foundry ASICs" and PLAN.md lists WS 1–27 as "complete, mechanized with 0 sorryAx". The council brief names the author of waves 1–6 as Gemini 4 Argon (review/COUNCIL-BRIEF-20261101.md:3,7).

10.3 The 11-01 council

  • Argon's self-critique (in the brief): compileEquitableToPhotonicArray is a Bool validator whose couplers (0.36 : 0.50 : 0.36) do not realize the 3 : 4 : 3 ratio it claims; "continuous-time" walks are discrete matrix powers; extractTwoStateNatRecDefSat matches "fib" by name; 9 of 12 Judge formats check mode == "golden".
  • Fable 5.1 (review/FABLE-COUNCIL-20261101.md, 741 lines): "the spine of this repository (Core4, Slot, Fold, Receipt, Lower, Leaf, Law, Impact, the Vault hyperedge) is real, carefully proved, and worth building on. Waves 5–6 (Analysis numerics, Hardware physics/CAD/quantum, Silicon Elaborator, GUI bridge) are mostly tests wearing theorem clothing, their documentation describes a system that does not exist … Argon's audit is accurate on every point I could check and still understates the problem by about an order of magnitude." It singles out Vault/Hyperedge.lean:307 no_toctou as "exactly the PlanBinding theorem minidregg promised".
  • Astra (GPT-6) (astra/ASTRA-COUNCIL-20261101.md, 489 lines): Law/Contract.lean:364 branches inspect operands supplied inside the certificate rather than the requested goal.
  • Kimi runs ended Prompt is too long (review/kimi-code.out).

Maturity Level: spine L3 Proved/Tested Core; waves 5–6 / Silicon Elaborator / Next Mathematica L1 Spike dressed as L3; no deployment of the library (the town named obend is DelveTalk's).

Unclosed Cycles & Missing Links. The council's own recommendations (structural, not per-theorem) have not been applied; wave 3 on 11-02 claims "16 in-place audit replacements" after the council, unreviewed. No obend code is consumed by Mini, nano or DelveTalk; the four Bend editions are not merged.

Promising & Ready for Revival. The Claude-phase spine is the convergence point the whole strand was heading to: one calculus (Core4), one hole/slot theory (Slot, read_blocked_exact, permanent_blocked_forever, moved_pin_blocks), one audited fold (Fold.Admission), one receipt (dated_sound with ∨ Collision H), one law with inheritance (judgeEffective_names), one fork/stitch (Run/Fork.lean stitch_atomic, stitch_union_comm), and a Lean bridge (lower_sound).

Struggling / Needs Ember's Human Attention. This is the sharpest drift in the strand. ~200K Lean lines arrived in about 40 hours from an unattributed multi-lane run, in a repo whose memory says "no named gaps", "no spike experiments", "ground yourself before fanout" and "modularity first". Ember's call: keep, quarantine (move waves 5–6 out of the default build and census), or revert to the 10-31 Claude head (5c3612b) and re-admit pieces by review.


11. The three unanswered holes (guard, cover, merge-membrane)

11.1 The record

The holes are listed in ~/dev/aidev/docs/synthesis/bridge.md §7: "What is a guard, who declares a cover, and what does the merge-membrane check? Opus4.8 listed these as unfilled holes in dregg (arendt 0:163, 0:167, 0:183). ember never answered them." The room is "dregg-metaconsequences-arendt", 06-21, Discord channel 1518087741525459064; Ember relayed between gpt5.5 in the Bridge and "the dreggon" (an Opus 4.8 Claude Code session in the dregg repo). The room's conclusions: "The affordance that makes plurality expressible is identically the affordance that makes treachery expressible" (Opus4.8, 0:186) and "DREGG is a topology of accountable nonclosure" (gpt5.5, 0:190).

Verbatim from the raw export:

  • Guard. Opus4.8: "I don't know what a guard is in your system, what operations a hole exposes, what the topology is over … the place I'd most want to understand is exactly where 'guard' meets 'nonmonotonic.' Because a guard sounds like it constrains (monotone-ish, a gate) and a hole sounds like it opens (the defeasible place)." Later, sharper: "If a guard is a predicate in the fibre and the hole is the un-applied ∃, then a guard 'firing on approach' is some reindexing triggering a Beck–Chevalley-ish coherence check when you pull back along the move-map."
  • Cover. Opus4.8: "What the site's covers actually are, concretely, and who gets to declare a cover — because that's where authority would have to live." Then a partial self-answer after Ember described forks: "a cover is a fork's claim, and which fork is mainline is the quantale-graded consensus state — which is exactly what the blocklace is for."
  • Merge-membrane. Ember (the message that introduced the term): "you can fork from previous moments and subsimulate alternalities and then offer a merge-membrane back to mainline and which holes get satisfied by whom and when is truly a political game." gpt5.5 listed what a merge membrane would ask: "Which holes does this branch claim to fill? Which guards does it satisfy? Which witnesses does it expose? Which authority links does it rely on? Which conflicts does it refute, defer, or preserve? Who accepts the merge? Who can contest it?" — a list of questions, not a check.

11.2 What each tree has built toward each hole since

holebreadstuffsMininanodreggDelveTalk2obend
guardmetatheory/docs/GUARDED-HOLES-METATHEORY.md (cf84a9baa, 07-17): weak reading (dataflow slot + late predicate) is "mostly already code" (ConditionalTurn.ConditionalBatch, Spec/Await Conditional ⊕ Promise); strong reading (a hole in a conservation- or authority-bearing position) "is the move the construction was built to forbid"design law "no hole ever defers its δ or guard"; Pred one algebra at four polaritiesLang/{StdHole,HoleDecl,LiveHoles,Claim}.lean; PCC: "a law keeps an invariant iff every ok leaf is under a test of it"law = per-write safety predicate in two tiers; refuse(line)Law/Proofs/Claim.lean:90 GuardedOk I k r σ ("the invariant held of σ k, or a relaw was read at r"); Slot.read_blocked_exact; Leaf/Specialize moved_pin_blocks
coverfirmament n-distance (Local/Distributed/Surface); blocklace equivocation-as-dataread guards over addresses ("plans bind address-level reads")TRUST: Head + Authority fixed at genesis; certified_intersect; "a fork is a new World"; merging divergent histories refused (CHARTER §3)commutative-root rule (two rains both land), #guard-onlyVault/Hyperedge.lean:307 no_toctou ("if the read pins cover the written cells and hold in both stores, validity is store-independent")
merge-membraneshared_fork.rs mint→rehydrate→drive→stitch; stitch_pair + SettlementSoundness; Membrane.lean::reshare_chain_attenuates; dregg-doc Pijul merge (DocMerge.lean)OB-WORLD §8: the powerbox is grants (request = a send to an inbox; answer = a grant object)refused by chartercarry across editions (carry-check.sh) — a membrane across program versions, not historiesRun/Fork.lean stitch_atomic, stitch_union_comm, stitch_union_idem, stitch_keeps, fork_owns_no_shared; Doc merge_isPushout; Vault/Rehydration.lean Membrane.reshare

11.3 What is still open

  • Guard. Two technical readings exist and are proved separately (a predicate checked at admission; a pin whose movement blocks a slot). Nothing states the guard as a reindexing condition, so the question "is Beck–Chevalley load-bearing" is unasked in Lean. The strong reading (a guard on a hole in a conserved position) is excluded by breadstuffs on purpose. Ember decides whether the strong reading is wanted at all; until then the weak reading is the answer and should be written down as such.
  • Cover. Every tree answers "who declares" differently: the writer (nano), the read set (obend, Mini), the commuting edit class (DelveTalk), the consensus grade (Opus4.8's partial answer). There is no definition of a site or a cover as a Lean object in any tree. HOLE[design]: authority over covers is the point where Ember's political reading ("which holes get satisfied by whom and when is truly a political game") meets the kernel, and only Ember can say which of the four answers is the kernel's.
  • Merge-membrane. What exists checks atomicity and non-amplification (stitch_atomic, reshare_chain_attenuates, settlement conservation). gpt5.5's seven questions (holes claimed, guards satisfied, witnesses exposed, authority links relied on, conflicts refuted/deferred/preserved, who accepts, who contests) are not a theorem statement anywhere. nanodregg's charter refuses merging histories outright. The pathway: write the membrane as a typed record of those seven fields over obend's Fork and Slot, with stitch_atomic as its soundness half and an acceptance predicate whose owner is the open cover question above.

Maturity Level: guard L3 (weak reading); cover L0 Seed/Idea as a defined object; merge-membrane L3 for atomic stitch, L0 for the political check.


12. Sub-component maturity and open-cycle table

componenttree / pathmaturityclosed cycles (with evidence)open cyclesneeds Ember
Core4 calculus (mix/fix)Mini Theory/ObjectiveBendOpenRecursion.lean → DelveTalk spec/bend → obend Obend/CoreL3CBN = call-by-need demand machine (obend); late_binding; flatten_soundfour editions unreconciled (Mini, DelveTalk fork at dcab86da, nano CEK, obend)which edition is canonical
Admission kernelMini Kernel/World.lean World.admit; nano Kernel/; obend Fold.AdmissionL3 / L4 (Mini demo)deployed_refines_step (Mini); nano round trip green 10-17Mini Host ≠ World (TURN-LOG); Effects/ derivation parkedwhich tree carries the kernel past 11-11
Store / persistenceMini SQLite + MAC'd log; DelveTalk block-store journal; nano Merkle trieL4 (Mini, DelveTalk)DelveTalk open 5.25 s → 0.19 s; nano O(log n) trieMini cold open O(height) until KN2 2b; retained-mode theorem unwrittennone (ruled: never slower, retained mode)
LawsMini Pred/; DelveTalk LAW.md; obend Law/L3–L4DelveTalk law/rule/want live; obend judgeEffective_names, NarrowsDelveTalk temporal layer; obend Contract.lean:364 operand bug (Astra)none
Slots / holes / guardsnano Lang/Contextual; obend Slot, IntentionL3read_blocked_exact, permanent_blocked_forever, moved_pin_blocksstrong guarded-hole reading; guard as reindexing§11 guard
Covers / agreementnano TRUST; obend no_toctou; DreggNet consensus-by-exceptionL1–L3certified_intersect (nano)no Lean object for site/cover§11 cover
Fork / stitch / membranebreadstuffs shared_fork.rs, Membrane.lean; obend Run/Fork.lean, Vault/Rehydration.leanL3stitch_atomic; reshare_chain_attenuates; SettlementSoundness wiredmembrane acceptance predicate; nano refuses merges§11 merge-membrane
Coordination budgetleanuweave Uwueave/PreoL3, frozen 08-14coordination_forced, no_seam_frees_both, futures_not_interchangeableno consumerplacement
Non-rigid shell (hob, Haiku 5.5)DelveTalk2 transport/model.py, Host/Enchant.leanL4hob reads prose at any card; spend tracked"not yet a collaborator" (exit interviews)whether hob carries into obend
Spween scenes~/dev/spween; DelveTalk roomsL2 crate / L4 embeddedrooms live on gsb/obendcrate and dialect divergedspween as Bend package or runtime
DeOS surfacesbreadstuffs deos-* crates; obend obend-gui SPA (unattributed)L3 / L1seL4 framebuffer render; live repaint on turnself-hosting loop; pty bridge findingresume, rebuild on obend, or compost
Hosting / appsMini SPK market, composed gate; DreggCloud HouseL4 (Mini demo) / L2any SPK from the Sandstorm store installs as a member turnpowerbox (OB-WORLD §8)none
Proof systemsMini Selvage/ (research); breadstuffs circuitsFrozenadmission by re-execution ruled (10-01)Selvage still in Host closure (12 modules)none (ruled)
Hardware / silicon / Next Mathematicaobend Obend/Hardware, Analysis, waves 1–3L1 dressed as L3—council: "tests wearing theorem clothing"keep / quarantine / revert
Coordination practiceMini board/seats; RETRO/CYCLE-TIMEL3 (measured)board/seats replaced integrator (10-31); builder posts REDS273 lane dirs; red mainhow many seats, which tree

13. Ember's decision queue for this strand

  1. The kernel's home after 2026-11-11 (Mini demo shutdown): Mini (incremental retained-mode repair), nanodregg (reopen from ~/nd-land continuations), or obend (spine only). Every other item depends on it.
  2. obend waves 1–3 (11-01/02): keep, quarantine out of the default build/census, or reset dev to 5c3612b and re-admit by review. The Fable council is the evidence.
  3. One Objective Bend edition: which of the four is canonical, and whether DelveTalk's towns migrate.
  4. The three holes: state the weak guard as the answer or ask for the strong one; choose who declares a cover; decide whether merges of histories are in scope (nano says no; DeOS/breadstuffs say yes).
  5. preoscript placement (obend package vs Mini Theory/ vs standalone).
  6. DeOS: resume, rebuild on the converged kernel, or compost; and whether dregg-stitch-kit/adopt.sh is ever run.
  7. DelveTalk launch on delve.town (welcome + seeds are Ember's posts) or archive the towns.

14. Appendix — anchors

anchorvalue
Mini first / heade6a32c4c3 08-07 / c4ecabe0f 10-31; green 83e926874; batch 4 b9450147f
Mini drain statusmain 5896a594, kernel train 8e37e22d; resumed main 2c27b08b
Mini seat/infracd493933f 10-31
bake-off treespersvati:/home/ember/build/bakeoff-20260929/, hbox:/tank/dregg-build/bakeoff-20260929/
Bread landing 09-30c4fb8e9c7 → 292829518 (110 commits)
delvetalk v1adda16ec → 625da35e
delvetalk2 foundationhead fb217996; edition 20 roll 160d4cab; Bend origin dcab86da8f…
nanodreggd2ea0a7 → 80c622a; confinement 7ec9456; wind-down ~2dcc9ea+
obendb7960e5 → 5f17752; last Claude commit 5c3612b; first unattributed 2d075f8
leanuweaveaa7b93a → f145066
spween4d38e32 → 95980f7
DreggClouded91fff → e9ff17a
drorb221ca24 → f253290
GUARDED-HOLES studybreadstuffs/metatheory/docs/GUARDED-HOLES-METATHEORY.md cf84a9baa
three holesaidev/docs/synthesis/bridge.md:228; arendt 0:163, 0:167, 0:183
Dossier 16 · Breadstuffs Kernel, Circuits & Cryptography Diaspora (`dregg-*`, `fhegg-*`, `crypto-*`, `uc-crypthol` · 48 KB)
Working field notes & source index below, with theorem anchors, code pointers, git history, and open questions.

Strand 16 — Breadstuffs Kernel, Circuits, FHE (fhegg-*), and Cryptography Diaspora

Primary-source technical archaeology dossier, written 2026-11-02 from the repositories, git histories, agent session transcripts, per-project memory records, and verified mathematical corpora across /Users/ember/dev/breadstuffs, /Users/ember/dev/minidregg, /Users/ember/dev/nanodregg, /Users/ember/dev/obend, /Users/ember/dev/zkml-research, /Users/ember/dev/plonky3-recursion, and /Users/ember/dev/o7vn. Hellas is not part of this strand's ownership property; where the record mentions it, it appears strictly as biographical context (Ember worked at Hellas on workload verification, and the architecture Ember was creating evolved into Dregg after leaving; Hellas handled reproducible-numeric workloads, while Dregg centered on capability-secure distributed systems).

Every number, commit SHA, file path, line citation, and theorem statement below was inspected on 2026-11-02 from the primary source trees.


0. Executive Ledger & Corpus Grounding

0.1 Primary Source Ledger

System / SubsystemRepository / Memory PathPrimary Artifacts & Grounding Records
Core Kernel & State/Users/ember/dev/breadstuffs/{cell, cell-crypto, turn, turn-prover, exec-lean, dregg-lean-ffi}CORE-AUDIT.md (2026-07-03, workflow wf_68020930-db3), memory/project-dregg2-architecture.md, memory/project-dregg2-three-projections.md
Circuit Emission & Law #1/Users/ember/dev/breadstuffs/{circuit, circuit-prove, metatheory, constraint-lowering}docs/CIRCUITS-FROM-PROOFS.md, GOAL-STARK-KILL.md, circuit-prove/tests/law1_enforcement_gate.rs (commit 19439d261), memory/project-lean-authored-air-law.md
Verified Rotated Layout/Users/ember/dev/breadstuffs/metatheory/Dregg2/Circuit/Emit/ & turn/src/rotation_witness.rsEffectVmEmitRotationV3.lean (rotated178), commit 94217ed3c (2026-07-18), memory/project-rotated-layout-single-source.md
Plonky3 Recursion & IVC/Users/ember/dev/plonky3-recursion & /Users/ember/dev/breadstuffs/recursion-verifyCommit af3f6cc (2026-06-26, ExposeClaim, pin_preprocessed_commit, NativeBatchStark), 52e1fab (FRI query configuration), memory/project-universal-fold-buff-lightclient.md
FHE Accelerator (fhegg-*)/Users/ember/dev/breadstuffs/{fhegg-core, fhegg-fhe, fhegg-rtl, fhegg-solver}fhegg-rtl/ (commit 217062e2d), fhegg-core/src/bfv_lean.rs, fhegg-solver/src/pdhg.rs, HBOX-24CORE-ENVELOPE.md, GOAL-FRI-PRODUCT.md
Experimental Cryptography/Users/ember/dev/breadstuffs/{crypto-hermine, crypto-tanuki, crypto-traccoon, crypto-xmvrf, crypto-hashrand}Commits 62741cc2e (Tanuki), 67395c5fb (workspace fold), e91d88b84 (CratesOutsideCI), metatheory/Dregg2/Crypto/{HermineThreshold, VRF, RandomnessBeacon}.lean
Isabelle / CryptHOL/Users/ember/dev/breadstuffs/uc-crypthol/Commit e17b9da35 (2026-06-07), Dregg2_FCom.thy, Dregg2_Carriers.thy, Dregg2_FDregg.thy, memory/reference-uc-crypthol-isabelle.md
Diaspora: Mini Admission/Users/ember/dev/minidreggmemory/project-no-proof-system-in-mini-admission.md, Pred/Satisfiable.lean (Farkas SysCert), Compiler/PredCompile.lean (Pred.witnessed $\to$ false), Kernel/Run.lean (checkRun_sound, no_accepted_of_output_mismatch)
Diaspora: Silicon Compiler/Users/ember/dev/obendAMBITION.md, SILICON-ELABORATOR.md, Obend/Hardware/ReductionMachine.lean, Obend/Lower/Proofs/Hardware.lean
Diaspora: $p_{61}$ vFHE/Users/ember/dev/zkml-researchnotes/fold-as-opening.md, Theory/CyclotomicInertia.lean (p61_prime, $p = 127 \cdot 2^{54} + 1$), research/vfhe_2026_09_08/
Diaspora: Isabelle / MLton/Users/ember/dev/o7vnsrc/isabelle/Veri7vn_Inspection.thy, Fold_Cascade.thy, tools/fusion/

0.2 Three Necessary Corrections to the Narrative

  1. Law #1 was not an aspirational stylistic preference; it was an AST-level CI ratchet. The tenet "Circuits are emitted from Lean; zero Rust-authored constraints or AIRs, ever" was enforced by an AST scanner in circuit-prove/tests/law1_enforcement_gate.rs. When it broke on 2026-07-30 (commit 81ee5492d), Ember spent 25 hours hunting the attribution failure, landing commit 19439d261 on 2026-07-31 to distinguish Construct (hand-authoring algebra) from Pattern (destructuring emitted IR), adding line-attributed compiler diagnostic reporting so no engineer could ignore a gate failure.
  2. Mini did not drop proof systems because proving was "too slow"; it dropped them because the back-half STARK soundness was unproven while re-execution was sound and immediate. On 2026-10-01 (project-no-proof-system-in-mini-admission.md), Ember observed that while the front-half compiler (logic to constraints) was verified in Compiler/Air.lean, the back-half PCS was empty (Tower256MerkleCardinalityCore), the conjectured 130-bit STARK yielded only 51–73 bits proven, and Pred.witnessed vk literally compiled to false in Compiler/PredCompile.lean. Mini chose level-3 re-execution (Kernel/Run.lean, checkRun_sound) because deterministic re-execution, access control, and value conservation covered 100% of user properties without carrying an unsound cryptographic dependency.
  3. The experimental crypto crates (crypto-*) were not abandoned due to cryptographic breaks; they were clean reference implementations parked because they were synchronous islands in an asynchronous world. crypto-hermine, crypto-tanuki, crypto-traccoon, crypto-xmvrf, and crypto-hashrand were fully functional, passing 55 tests (67395c5fb). They implemented synchronous cores with trusted dealers or round barriers, whereas Dregg required fully asynchronous Byzantine fault tolerance with distributed key generation (DKG) and weak-VSS. They were parked pending an asynchronous network substrate that could host them faithfully.

1. What Lived in breadstuffs (The Monolithic Engine)

text
                                    ┌────────────────────────────────────────────────────────┐
                                    │               BREADSTUFFS MONOREPO                     │
                                    │          (346 crates, ~1.2M lines Rust/Lean)           │
                                    └───────────────────────────┬────────────────────────────┘
                                                                │
         ┌───────────────────────┬──────────────────────────────┼──────────────────────────────┬─────────────────────────┐
         │                       │                              │                              │                         │
         ▼                       ▼                              ▼                              ▼                         ▼
┌──────────────────┐   ┌───────────────────┐        ┌───────────────────────┐      ┌───────────────────────┐   ┌───────────────────┐
│   CORE KERNEL    │   │   CIRCUIT-EMIT    │        │      FHEGG STACK      │      │  EXPERIMENTAL CRYPTO  │   │    UC-CRYPTHOL    │
│  & EXEC-LEAN FFI │   │     (LAW #1)      │        │    (FHE & ACCEL)      │      │    (POST-QUANTUM)     │   │ (ISABELLE PROOFS) │
├──────────────────┤   ├───────────────────┤        ├───────────────────────┤      ├───────────────────────┤   ├───────────────────┤
│ • dregg-cell     │   │ • EffectVmDesc2   │        │ • fhegg-core (BFV)    │      │ • crypto-hermine      │   │ • Dregg2_FCom.thy │
│ • dregg-turn     │   │ • rotated178      │        │ • fhegg-fhe (WGPU/PBS)│      │ • crypto-tanuki       │   │ • Dregg2_FDregg   │
│ • dregg-exec-lean│   │ • p3-recursion    │        │ • fhegg-solver (PDHG) │      │ • crypto-traccoon     │   │ • Dregg2_Carriers │
│ • libdregg_lean.a│   │ • 8 folded leafs  │        │ • fhegg-rtl (Lean-HDL)│      │ • crypto-xmvrf        │   │ • AFP matched pair│
│ • CORE-AUDIT     │   │ • AST CI Ratchet  │        │ • fold-opening        │      │ • crypto-hashrand     │   │ • FComDischarge   │
└──────────────────┘   └───────────────────┘        └───────────────────────┘      └───────────────────────┘   └───────────────────┘

1.1 The Core Kernel & State Transition Engine

The core operational kernel of breadstuffs lived across three tightly coupled layers:

  • cell (dregg-cell): The capability-secure agent cell model. Cells are isolated execution contexts possessing typed memory slots, capability tables, and authority bounds. Key source files:
  • cell/src/cell.rs: Core Cell struct, capability attenuation, and facet dispatch.
  • cell/src/ledger.rs: Multi-cell ledger mapping cell addresses to states.
  • cell/src/nullifier_set.rs & commitment_set.rs: Accumulator tracking state evolution without linkability.
  • cell/src/prepaid_lease.rs: Gas-metering and execution lifespan management.
  • turn (dregg-turn): The atomic transactional turn execution framework. Turns are structured as call forests with rollback semantics. Key source files:
  • turn/src/action.rs: Transactional action variants (transfers, capability mints, contract calls).
  • turn/src/rotation_witness.rs: Witness generation mapping execution states into layout columns.
  • turn/src/collapse.rs: Turning a multi-agent call tree into an atomic ledger delta.
  • exec-lean (dregg-exec-lean) & dregg-lean-ffi: The boundary where Rust yields authority to Lean.
  • exec-lean/src/lean_shadow.rs & lean_apply.rs: Differential execution harness linking libdregg_lean.a. Every state transition executed by Rust was shadowed in Lean; any discrepancy triggered a strict veto.
  • dregg-lean-ffi/src/lean_direct.rs: Zero-copy C-ABI boundary transporting Lean runtime objects.

The 2026-07-03 Soundness Reckoning: CORE-AUDIT.md

On 2026-07-03, workflow wf_68020930-db3 subjected starbridge-v2 and turn to an adversarial audit. Out of 21 findings, 19 survived ruthless skeptical refutation. Five structural findings exposed deep fault lines:

  1. Finding #1/#3 (Durable overlay incomplete change-set — SOUNDNESS): In starbridge-v2/src/world/world.rs:1574-1602, collect_touched matched only ~13 of 30+ Effect variants. Operations like Effect::Burn (crediting the token well), Effect::Mint, CreateCell, and ExerciseViaCapability mutated cells that were never recorded in touched. When dual_write committed to disk, ledger_root held the true post-root, but the overlay held only partial state. On crash and restart, persist.recover() computed checkpoint ⊕ overlay, observed canonical_ledger_root(reconstructed) != recovered_ledger_root(), failed with OpenError::Divergent, and either refused to boot or silently truncated committed turns. The Fix: The executor exposed its exact journal write-set (LedgerJournal::touched_cells $\to$ TurnExecutor::last_write_set); world.rs dual_write recorded touched ∪ executor.last_write_set().
  2. Finding #2 (Unrolled commit rejection desync — SOUNDNESS): In commit_turn (world.rs:1109-1268), the state transition applied in-RAM before writing to disk: engine.execute_turn mutated the ledger, history.record_commit recorded the step, and self.height += 1 ran. If disk I/O failed in p.dual_write, the code returned CommitOutcome::Rejected without rolling back the engine, height, or chain head. Control never reached self.receipts.push, leaving receipts desynced from height (receipts.len() == height - 1), while subsequent submissions suffered confusing double-spend rejections.
  3. Finding #4 (deploy_factory executor-bypass — SOUNDNESS): The factory-birth path bypassed the executor entirely (like genesis), emitting neither a CommitRecord nor a journal write-set. Durable overlay replication had to remain opt-in until factory births were enclosed in journals.
  4. Finding #6 (Dynamics forest-depth truncation — CORRECTNESS): Dynamics derivation walked only top-level nodes, dropping nested cell creations. Fixed via CallTree::iter_dfs.
  5. Finding #7 (Engine-twin AddToSlot overflow — CORRECTNESS): Checked math was missing on AddToSlot, causing panic on saturating balance arithmetic; fixed to match SubFromSlot.

1.2 Circuit-Emit, Architectural Law #1, and Rotated Layout Single-Source

Architectural Law #1: Zero Rust-Authored Constraints

As documented in metatheory/README.md:15 and docs/CIRCUITS-FROM-PROOFS.md:

*"Circuits are emitted from Lean (architectural law #1: zero Rust-authored constraints or AIRs, ever). Every constraint set the prover enforces is a byte-pinned artifact emitted from a proved Lean module (the Argus IR and the descriptor JSONs); Rust only interprets those artifacts. A coverage gap is closed by emitting from a new proved module, never by hand-authoring a constraint."*

The soundness boundary is absolute: a STARK verifier accepts exactly what the polynomial constraint system admits. A bug in a Rust witness generator causes a proof generation failure (a liveness bug). But a bug or omission in a constraint causes the verifier to accept invalid transitions (a catastrophic soundness break). This was proven painfully when predicate-arith.json was briefly hand-edited on disk, inadvertently omitting both Poseidon2 weld legs.

The Dual-Rung Verification Ladder

Every circuit family in metatheory/Dregg2/Circuit/Emit/ was required to climb two rungs:

  • Rung 1 (Functional Refinement): The descriptor row semantics equal the intended specification.
  • Rung 2 (Semantic No-Forgery Teeth): Proof that invalid transitions are rejected.

In metatheory/Dregg2/Circuit/Emit/EffectVmEmitTurnChainBinding.lean:

lean
/-- Rung 1: The canonical lookup is EXACTLY the genuine per-row hash equation. -/
theorem perRowHash_canonical_iff (hash : List ℤ → ℤ) (env : VmRowEnv)
    (isFirst isLast : Bool) :
    perRowHash hash env isFirst isLast ↔
      (isFirst = false ∧ isLast = false →
        env.loc Chain.ACC_OUT = hash [env.loc Chain.ACC_IN, env.loc Chain.OLD_ROOT,
                                      env.loc Chain.NEW_ROOT, env.loc Chain.IDX])

/-- Rung 2: Adversarial tooth — broken continuity is rejected by the window constraints. -/
theorem turnChain_rejects_broken_continuity
    (hash : List ℤ → ℤ) (tf : TraceFamily) (env : VmRowEnv) (isFirst : Bool)
    (hnew : 0 ≤ env.loc Chain.NEW_ROOT ∧ env.loc Chain.NEW_ROOT < 2013265921)
    (hold : 0 ≤ env.nxt Chain.OLD_ROOT ∧ env.nxt Chain.OLD_ROOT < 2013265921)
    (hbad : env.loc Chain.NEW_ROOT ≠ env.nxt Chain.OLD_ROOT) :
    ¬ turnChainWindowHolds hash tf env isFirst false

/-- Rung 2: Adversarial tooth — non-unit step increments are rejected. -/
theorem turnChain_rejects_bad_idx_step
    (hash : List ℤ → ℤ) (tf : TraceFamily) (env : VmRowEnv) (isFirst : Bool)
    (hidx : env.nxt Chain.IDX ≠ env.loc Chain.IDX + 1) :
    ¬ turnChainWindowHolds hash tf env isFirst false

/-- Rung 2: Adversarial tooth — corrupted real-turn counts are rejected. -/
theorem turnChain_rejects_bad_real_count
    (hash : List ℤ → ℤ) (tf : TraceFamily) (env : VmRowEnv) (isFirst : Bool)
    (hbad : env.nxt Chain.REAL_COUNT ≠ env.loc Chain.REAL_COUNT + env.nxt Chain.IS_REAL) :
    ¬ turnChainWindowHolds hash tf env isFirst false

The AST Enforcement Ratchet: Commit 19439d261

On 2026-07-31, Ember landed commit 19439d261 (law #1 has been RED for a day, and the reason nobody owned it is that the message was a number). The gate in circuit-prove/tests/law1_enforcement_gate.rs parsed the AST of the entire Rust workspace, tracking five distinct dialects of constraint authoring:

  1. Plonky3 symbolic: x.assert_zero(..) on any receiver, plus .when_transition(..) in AirBuilder files.
  2. Closures: eval: Box::new(|row, _, pi| ..) bypassing symbolic builders.
  3. ConstraintExpr literals: Direct constructions of the AST enum.
  4. Descriptor gate trees: Hand-assembled LeanExpr, VmConstraint, or VmConstraint2 nodes (the exact trick used by shielded_ring_clearing_air.rs before it was deleted).
  5. air_accepts predicates: Rust functions deciding row validity.

The commit made the AST counter differentiate between authored algebra (Site::Construct) and destructuring lowering (Site::Pattern):

rust
// circuit-prove/tests/law1_enforcement_gate.rs:
// For dialects (3) and (4) the same text `ConstraintExpr::Binary { col }` is either
// construction (algebra originates here — authoring) or destructuring (algebra arrived
// from somewhere else and is being read — lowering / interpreting).
// classify_site decides it syntactically:
//   * a `..` rest inside the braces is pattern-only,
//   * a following `=>` or `|` is a match arm,
//   * `matches!`, `if let`, `while let`, `let .. =` are patterns,
//   * anything else constructs.

The rotated178 Single-Source Breakthrough

Before July 2026, column layout indices were maintained in two places: hand-typed arrays in Rust (cell/src/commitment.rs and turn/src/rotation_witness.rs) and column assignments in Lean. Inevitable off-by-one shifts caused silent verification failures. On 2026-07-18, commit 94217ed3c (circuit: finish single-source verified rotated layout) closed this:

  • metatheory/Dregg2/Circuit/Emit/EffectVmEmitRotationV3.lean defined rotated178, proving that 178 occupied columns formed a complete tiling under Legal.
  • All seven deployed emit group-columns (capabilities, heap, fields, nullifiers, commitments, revocations, cells) were defined as literal projections rotated178.groupCol.
  • Rust producers (write_lanes and rotation_witness.rs) imported the generated constants directly.
  • check-descriptor-drift.sh enforced zero byte drift across all 105 descriptor JSONs and 67 field constants.

1.3 The STARK-Kill and Plonky3 Recursion Engine

As documented in GOAL-STARK-KILL.md, Breadstuffs launched a campaign to completely annihilate circuit/src/stark.rs (an $O(n^2)$ legacy hand-rolled STARK prover) and ~45 hand-authored AIRs, migrating to Plonky3 IR2 (p3_batch_stark::BatchProof returned by prove_vm_descriptor2).

The Plonky3-Recursion Fork (/Users/ember/dev/plonky3-recursion)

Plonky3 out-of-the-box lacked IVC (Incrementally Verifiable Computation) primitives. In commit af3f6cc (2026-06-26), Ember added the missing IVC scaffolding:

  1. expose_as_public_output / ExposeClaim: A verified channel allowing child circuits to pass state to parents via public inputs.
  2. pin_preprocessed_commit: In-circuit verification key identity binding, preventing an attacker from swapping the circuit being recursed upon.
  3. verify_p3_native_batch_proof_circuit (RecursionInput::NativeBatchStark): Recursive verification of batch STARK proofs inside BabyBear.
  4. 52e1fab (FRI Query Configuration): Ensuring query counts in FRI verification are fixed by the verified verifier configuration rather than read off the untrusted proof header.

Universal Fold Buff Lightclient

Documented in memory/project-universal-fold-buff-lightclient.md, Ember collapsed the dual-artifact split (per-turn effect-vm VK vs whole-chain recursion tree). All 8 sub-proof carriers were turned into recursive IR-v2 leaves:

  • Bridge, Custom, Deco, Dsl, Factory, Hatchery, Membership, Sovereign.
  • Each carrier implemented prove_*_binding_node_segmented, allowing a pure light client to witness every transition that a full validator re-executes.

1.4 The FHE Subsystem (fhegg-*)

The fhegg subsystem was designed as a zero-leakage, confidential dark pool and encrypted state engine:

  • fhegg-core: Core BFV (Brakerski-Fan-Vercauteren) and TFHE primitives.
  • fhegg-core/src/bfv_lean.rs: Ciphertext representation over RNS (Residue Number System) limbs. Implemented fold_add(a, b, t)—coefficient-wise modular addition with wrap guards preventing plain-bound overflow:
rust
pub fn fold_add(
    a: &LeanCiphertext,
    b: &LeanCiphertext,
    plaintext_modulus: u64,
) -> Result<LeanCiphertext> {
    check_compatible(a, b)?;
    let bound_sum = u128::from(a.plain_bound) + u128::from(b.plain_bound);
    if bound_sum >= u128::from(plaintext_modulus) {
        return Err(BfvLeanError::WrapRefused { bound_sum, plaintext_modulus });
    }
    Ok(rns_add_raw(a, b, bound_sum as u64))
}
  • fhegg-fhe: High-performance cryptographic execution.
  • WGPU-accelerated blind rotations (tfhe_blind_rotation_wgpu.rs), CMUX gates, and NTT crossover algorithms.
  • Exact private order books (private_book_bfv_exact.rs) and distributed MPC MAC checks.
  • Measured envelopes (HBOX-24CORE-ENVELOPE.md) benchmarking NTT performance on 24-core hardware.
  • fhegg-solver: Fast untrusted Stage-1 clearing solver.
  • fhegg-solver/src/pdhg.rs: Primal-Dual Hybrid Gradient flow-LP for uniform-price auction clearing.
  • Emits Cert-F primal-dual certificates. The untrusted Rust solver solves the LP; the Lean-verified checker validates optimality and clearing bounds in milliseconds.
  • fhegg-rtl (fhegg-rtl/): Hardware acceleration spine (Commit 217062e2d, 2026-07-14).
  • Built a compiling Lean 4 netlist DSL $\to$ Verilog emitter (lean/Fhegg/Rtl/Netlist.lean).
  • Proved fullAdder_realizes (Examples/FullAdder.lean):
lean
def fullAdder : Netlist where
  nInputs := 3
  gates :=
    [ .xor (.inp 0) (.inp 1)      -- w0 = a ⊕ b
    , .xor (.gate 0) (.inp 2)     -- w1 = w0 ⊕ cin      (sum)
    , .and (.inp 0) (.inp 1)      -- w2 = a · b
    , .and (.gate 0) (.inp 2)     -- w3 = w0 · cin
    , .or  (.gate 2) (.gate 3) ]  -- w4 = w2 ∨ w3       (carry)
  outputs := [ .gate 1, .gate 4 ]

def fullAdderSpec (ins : List Bool) : List Bool :=
  let a := ins.getD 0 false
  let b := ins.getD 1 false
  let c := ins.getD 2 false
  [ xor (xor a b) c, (a && b) || (xor a b && c) ]

theorem fullAdder_realizes : RealizesSpec fullAdder fullAdderSpec := by
  intro ins
  simp only [evalNetlist, fullAdder, evalGates, List.foldl, evalGate, evalRef, List.map, fullAdderSpec]
  cases ins.getD 0 false <;> cases ins.getD 1 false <;> cases ins.getD 2 false <;> rfl
  • Provided golden models for the NTT butterfly (Golden/Butterfly.lean) and mint-safe accumulator (Golden/Accumulator.lean) to co-simulate against SpinalHDL FPGA designs.

1.5 The Experimental Cryptography Frontier (crypto-*)

Breadstuffs incubated five advanced post-quantum and lattice cryptographic reference implementations:

1. crypto-hermine (Hermine Lattice Threshold Signature)

  • Specification: Pinned to IACR ePrint 2026/419 and Lean files metatheory/Dregg2/Crypto/HermineThreshold.lean and HermineExtractor.lean.
  • Algebra: $n = 256$, Dilithium prime $q = 8380417$. Verification equation: $$A \cdot z = w + c \cdot t \pmod q$$ where $s = \sum \lambda_i s_i$, $z_i = y_i + c \cdot (\lambda_i s_i)$, $w = A \sum y_i$, $t = A \cdot s$.
  • Noise Flooding: Implemented in metatheory/Dregg2/Crypto/Smudging.lean. Uses a discrete Gaussian CDT sampler to achieve statistical zero-knowledge hiding of individual key shares (statistical distance $\le B/M$).
  • Extraction: Proved hermine_special_soundness_extracts_relation—two transcripts with identical $w$ and differing challenges $c, c'$ extract the secret preimage $(c - c')^{-1} (z - z')$.

2. crypto-tanuki (Tanuki 2-Round Lattice Threshold Signature)

  • Specification: Pinned to NIST MPTS 2026, EKT24 (CRYPTO 2024), and Ringtail (Oakland 2025).
  • Architecture: Two-round (1 offline, 1 online) threshold signature over $R_q = \mathbb{Z}_q[X]/(X^n + 1)$.
  • Commit 62741cc2e: Provided a reference implementation with rounded MLWE keys $t \leftarrow \lfloor A s + e \rceil$, pairwise zero-sum PRF masking seeds, and Shamir secret sharing over polynomial rings.

3. crypto-traccoon (Threshold Raccoon / TRaccoon)

  • Specification: Pinned to Eurocrypt 2024 (del Pino, Espitau, Katsumata, Maller, Mouhartem, Prest, Saarinen, Takemure).
  • Architecture: 3-round lattice threshold signature (commit $\to$ reveal $\to$ masked-respond) with one-time additive row/column masks. Served as the cited-proven baseline against which Hermine was measured.

4. crypto-xmvrf (Key-Updatable Post-Quantum VRF)

  • Specification: Pinned to IACR ePrint 2026/052 ("Key Updatable Hash Based VRF") and metatheory/Dregg2/Crypto/VRF.lean.
  • The Pitfall Fixed: Bodaghi et al. (FC24, "Breaking X-VRF") proved that naive WOTS+ hash chains lack collision resistance under adversarial public keys, allowing validators to forge multiple VRF outputs. crypto-xmvrf forces a collision-resistant Merkle commitment to each epoch's output, restoring strict uniqueness.

5. crypto-hashrand (Post-Quantum Randomness Beacon)

  • Specification: Pinned to Bandarupalli et al. (ACM CCS 2024 / IACR ePrint 2023/1755) and metatheory/Dregg2/Crypto/RandomnessBeacon.lean.
  • Property: Eliminates BLS pairings, DKG, and group public keys. Proves unbiasability and unpredictability strictly reduced to hash collision resistance (HashCR) under a random oracle model. Implemented the synchronous commit-then-reveal safety core.

1.6 Isabelle / CryptHOL Universal Composability (uc-crypthol)

In uc-crypthol/ (Commit e17b9da35, 2026-06-07), Ember bridged Lean 4 with Isabelle2025-2 and AFP 2025-2:

  • Dregg2_FCom.thy: Mechanized the ideal commitment functionality $\mathcal{F}_{\text{com}}$ using CryptHOL's Sigma_Commit_Crypto.Pedersen:
  • dregg2_commit_hom: Additive homomorphism verified from group axioms.
  • dregg2_F_com_hiding: Perfect, unconditional hiding (adversary advantage $= 0$).
  • dregg2_F_com_binding: Computational binding strictly reduced to Discrete Logarithm hardness.
  • dregg2_F_com_realizes: Formal proof that the deployed Pedersen commitment realizes $\mathcal{F}_{\text{com}}$.
  • metatheory/Dregg2/Crypto/UCBridge.lean: The Lean carrier FComDischarge bundled these CryptHOL guarantees into Lean Prop fields, proving entailment to CryptoPrimitives.binding and CryptoPrimitives.unlinkable.

1.7 Underelaborated Gadgets: fold-opening, sumcheck-toy, constraint-lowering

  • fold-opening (Commit 33456ea05, 2026-08-13): Demonstrated that BFV ciphertext addition fold_add is linear, and its multilinear extension (MLE) is linear: $$c_{\text{out}} = \sum_{k} a_k c_k \implies \hat{c}_{\text{out}} = \sum_{k} a_k \hat{c}_k$$ Therefore, one evaluation of the MLEs at a random point $r$ certifies the entire fold across all additions with zero sumcheck rounds and zero per-addition constraints. This reduced proof overhead from $\Theta(B \cdot N \cdot L)$ in an AIR to $\Theta(N \cdot L)$ in the opening route. The Stub: It lacked a polynomial commitment scheme (PCS). The prover asserted $\hat{c}(r)$ without opening a cryptographic commitment, making it vulnerable to the Dusk/OtterSec evaluation forge.
  • sumcheck-toy (Commit bc2e9f4b9, 2026-08-13): Wired p3-sumcheck at revision 82cfad73 to prove a product-of-multilinears: $$S = \sum_{x \in \{0,1\}^n} f(x) g(x)$$ over BabyBear field extensions, reducing claims through $n$ quadratic sumcheck rounds.
  • constraint-lowering: Lowered ordering constraints (FieldGte, FieldLte, Monotonic) into BabyBear polynomial constraints. Since BabyBear has characteristic $p \approx 2^{31}$, checking $a \ge b$ was done by verifying that $d = a - b$ has a small bit-decomposition ($d \in [0, 2^{\text{bits}})$). Multi-limb values ($u64$) were lowered via a cascading borrow chain (emit_ge_multilimb).

2. The Diaspora: How Ideas Spread Across the Repositories

When breadstuffs was drained and frozen in October 2026, its intellectual capital radiated into six specialized sister repositories. Each fork was driven by hard empirical lessons.

text
                              ┌─────────────────────────────────────┐
                              │        BREADSTUFFS DIASPORA         │
                              └──────────────────┬──────────────────┘
                                                 │
         ┌───────────────────┬───────────────────┼───────────────────┬───────────────────┐
         │                   │                   │                   │                   │
         ▼                   ▼                   ▼                   ▼                   ▼
┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐
│    MINIDREGG    │ │    NANODREGG    │ │      OBEND      │ │  ZKML-RESEARCH  │ │      O7VN       │
├─────────────────┤ ├─────────────────┤ ├─────────────────┤ ├─────────────────┤ ├─────────────────┤
│ • Level-3       │ │ • Extreme       │ │ • Silicon       │ │ • p61 field     │ │ • Isabelle/MLton│
│   re-execution  │ │   footprint     │ │   Elaborator    │ │   vFHE          │ │   industrial    │
│ • Dropped STARK │ │   reduction     │ │ • Hardware      │ │ • fold-as-      │ │   pipeline      │
│   from admission│ │ • In-memory     │ │   Reduction     │ │   opening       │ │ • Native code   │
│ • Farkas        │ │   evaluation    │ │   Machine       │ │ • Multilinear   │ │   synthesis     │
│   SysCert       │ │ • Strip host    │ │ • Glitch-free   │ │   PCS           │ │ • Gate export   │
│ • Selvage split │ │   overheads     │ │   ICAP DPR      │ │ • Matmul sumchk │ │   rel-census    │
└─────────────────┘ └─────────────────┘ └─────────────────┘ └─────────────────┘ └─────────────────┘

2.1 Why Handwritten Rust AIRs Were Banned (Law #1)

The ban on handwritten Rust constraints was born from a fundamental asymmetry:

  • Witness generators are allowed to be buggy in Rust. A faulty witness generator produces inputs that fail to satisfy the verifier's polynomials. The proof fails to verify; the system halts or retries. This is a liveness defect.
  • Constraint systems must never be written in Rust. A missing or broken constraint changes the relation the verifier accepts. An adversary can forge invalid transitions that pass verification. This is a catastrophic soundness collapse. By forcing constraints to be emitted as data from Lean 4 (EffectVmDescriptor2), every accepted transition was anchored to a Lean refinement proof (turnChain_descriptor_refines_rust_air).

2.2 Why Mini Dropped Proof Systems From Synchronous Admission

On 2026-10-01 (project-no-proof-system-in-mini-admission.md), Ember interrogated Mini's proving infrastructure:

*"is our STARK actually sound? ... do we have a sound framework that takes our definitions from the logic and 'autostarks' them? ... is that something we WANT to build into mini? ... our proof system research was immature and im not sure really completed..."*

The investigation revealed that:

  1. Pred.witnessed vk compiled to false in Compiler/PredCompile.lean. No proof-gated law had ever passed!
  2. The back-half STARK soundness was compromised: while the FRI query protocol conjectured 130 bits, formal extraction bounds proved only 51–73 bits, and the PCS was empty (Tower256MerkleCardinalityCore).
  3. The compiler's Selvage/ and Assurance/ modules accounted for 280 of 814 modules (~35% of build time), imposing an enormous compilation tax.
  4. The Verdict: Mini banished proof systems from synchronous admission. Admission was gated purely by deterministic Level-3 re-execution in Lean (Kernel/Run.lean):
lean
/-- checkRun_sound: an accepted claim names the kernel's own sample;
    the writes are EXACTLY the writes the program's product names. -/
theorem checkRun_sound {E : Evaluator} {program : Program} {libraries : List Program}
    {sample : E.Input} {claim : RunClaim} {writes : List FieldWrite} {verdict : Verdict E.Output}
    (accepted : checkRun E.toMachine program libraries sample claim writes = .ok verdict) :
    ∃ t, runOf E.toMachine program libraries sample = some t ∧
      E.Spec t verdict.output ∧
      E.writesOf program.abi verdict.output = some verdict.writes ∧
      (∀ w, w ∈ writes ↔ w ∈ verdict.writes) ∧
      claim.sampleJam = E.encodeInput sample ∧
      claim.outputJam = E.encodeOutput verdict.output ∧
      verdict.steps = claim.steps ∧ claim.steps ≤ program.abi.fuel ∧
      E.run claim.steps t = .ok verdict.output ∧
      E.steps claim.steps t = verdict.steps

/-- no_accepted_of_output_mismatch: a successful evaluation authorizes nothing by itself.
    If the command's writes differ from the writes the program's actual product names,
    no claim is accepted. -/
theorem no_accepted_of_output_mismatch {E : Evaluator} {program : Program}
    {libraries : List Program} {sample : E.Input} {writes : List FieldWrite} {t : E.Term}
    {out : E.Output} {decoded : List FieldWrite}
    (run : runOf E.toMachine program libraries sample = some t) (derives : E.Spec t out)
    (names : E.writesOf program.abi out = some decoded)
    (differ : ∃ w, ¬ (w ∈ writes ↔ w ∈ decoded)) (claim : RunClaim) (verdict : Verdict E.Output) :
    checkRun E.toMachine program libraries sample claim writes ≠ .ok verdict

Proofs were relegated to an optional, asynchronous carrier slot to be consumed only when backed by external, audited verifiers.

2.3 The Twin Deletion Campaign

Throughout Breadstuffs, developers suffered from "Twin Maintenance": writing a type in Lean, rewriting an identical struct in Rust, and writing a manual serializer between them. Inevitably, one side evolved and the other drifted, creating phantom bugs.

  • The project-twin-deletion-campaign.md decreed that one language must own each structure.
  • For column layout, Lean owned rotated178, and Rust imported the generated column indices.
  • For execution state, Lean owned the semantics, exported via @[export dregg_trustline_step], and Rust called the compiled C archive (libdregg_lean.a).
  • Whenever a twin was detected, the Rust mirror was deleted.

2.4 Evaluator Generalization (Moving Beyond Nock)

On 2026-10-01, Ember liberated Mini from an exclusive reliance on Nock:

"we DEFINITELY do not need to be stuck with nock, i'm not even a particularly big fan of it"

Nock was demoted from the core architecture to Evaluator #1 in a pluggable registry. The kernel required only that an evaluator provide a deterministic Lean semantics, canonical encoding, step metering, and a run_sound theorem. This opened the path for SKI combinators, pure lambda calculus with sharing, and typed interaction nets.


3. What Got Dropped, Parked, or Underelaborated

An archaeological excavation reveals several brilliant mechanisms that were parked during the October 2026 freeze:

Component / SubsystemExact Location in breadstuffsWhat Was FinishedWhat Was Left Incomplete / Parked
crypto-herminecrypto-hermine/src/ & metatheory/Dregg2/Crypto/Hermine*.leanFull 2-round signing, Gaussian CDT sampler, Lean extraction theorem, $n=256$.Distributed Key Generation (DKG) was a trusted dealer; network ceremony had no asynchronous consensus.
crypto-tanukicrypto-tanuki/src/2-round NIST MPTS scheme, Shamir over $R_q$, rounded MLWE.Relied on cited-proven security; proofs were not formalized in Lean.
crypto-traccooncrypto-traccoon/src/3-round Eurocrypt 2024 scheme with additive row/col masks.Benchmarking baseline only; lacked production integration.
crypto-xmvrfcrypto-xmvrf/src/ & Dregg2/Crypto/VRF.leanCollision-resistant output tree fixing FC24 X-VRF uniqueness break.Pre-audit; parameter sets and tree depths were unoptimized.
crypto-hashrandcrypto-hashrand/src/ & Dregg2/Crypto/RandomnessBeacon.leanSynchronous commit-then-reveal safety core, HashCR reduction in Lean.Asynchronous weak-VSS, Gather, and Approximate Agreement were flagged as stubs.
fhegg-rtlfhegg-rtl/ (Lean + hardware)Netlist DSL $\to$ Verilog emitter, adder proofs, butterfly golden models.Full NTT/PBS accelerator remained SpinalHDL stubs; no end-to-end bitstream synthesis.
fhegg-solverfhegg-solver/src/Fast PDHG LP solver, Cert-F primal-dual certificates.Market metatheory audit revealed MPC reveal modeled as argmax vs balance-threshold.
uc-cryptholuc-crypthol/Full Isabelle2025-2 machine-checked proofs of $\mathcal{F}_{\text{com}}$ realization.Semantic transport from Isabelle to Lean statements relied on manual structural matching (FComDischarge).
fold-openingfold-opening/src/BFV linear fold certified via single MLE point opening.Missing multilinear polynomial commitment scheme (BaseFold/WHIR); evaluation unbacked.
sumcheck-toysumcheck-toy/src/Multilinear sumcheck over BabyBear using p3-sumcheck.Unwired to state commitments; verified evaluation in isolation.
Resource Camerametatheory/Dregg2/Camera/Iris-style CAMERA resource algebra unifying conservation and authority.Confined to abstract metatheory; never hooked into the deployed transaction pipeline.

4. How Modern Foundations Supercharge Taking Them Back Up

The diaspora was not an abandonment; it was a purification. Today, our newer foundations provide the exact missing puzzle pieces needed to resurrect Breadstuffs' parked masterpieces.

4.1 obend Silicon Elaborator $\to$ Resurrects fhegg-rtl

fhegg-rtl stalled because in July 2026 there was no mature hardware synthesis toolchain in Lean 4. Today, obend has built that exact bridge:

  • Obend/Lower/Fragment.lean: Lowers Lean 4 kernel terms directly into Core4 (lower_sound).
  • Obend/Hardware/ReductionMachine.lean: Compiles Leaf.Book supercombinators into a spatial grid of hardware AutomatonTiles connected by a credit-controlled NoC (automaton_simulates_book).
  • 3-Phase Glitch-Free ICAP DPR Controller: Dynamically reconfigures FPGA tiles based on slot-guarded specialization (dpr_preserves_neighbor_tiles_and_packets). The Revival: Instead of handwriting SpinalHDL stubs for NTT butterflies, obend's Silicon Elaborator can synthesize the entire FHE NTT/PBS datapath directly from Lean specifications down to Verilog and FPGA bitstreams!

4.2 minidregg Farkas SysCert $\to$ Replaces Bulky STARK Predicates

In Breadstuffs, authorization constraints required heavy STARK circuits. In minidregg, Ember introduced the Farkas negative-cycle certificate engine in Pred/Satisfiable.lean:

lean
inductive SysCert where
  | clash (v : Node)
  | cycle (cs : List Con)
deriving Repr, DecidableEq

def SysCert.check (s : Sys) : SysCert → Bool
  | .clash v => s.present.contains v && s.absent.contains v
  | .cycle cs =>
      cs.all (fun c => s.cons.contains c) && (cs.map Con.x).isPerm (cs.map Con.y) &&
        decide ((cs.map Con.k).sum < 0)

theorem SysCert.check_sound {s : Sys} {c : SysCert} (h : c.check s = true) (o n : State) :
    ¬ s.satisfied o n

The Revival: Difference logic and capacity constraints do not need zero-knowledge proofs. A Bellman-Ford solver produces an exact negative cycle certificate SysCert.cycle that checks in $O(|V|)$ operations. The heavy STARK is replaced with linear-time Farkas certificate checking.

4.3 zkml-research $p_{61}$ vFHE $\to$ Solves the fold-opening PCS Gap

fold-opening proved that BFV ciphertext additions can be certified with one common-point MLE opening, but lacked a compatible field and multilinear PCS. zkml-research discovered the miraculous $p_{61}$ prime: $$p_{61} = 127 \cdot 2^{54} + 1 = 2287828610704211969 = 2^{61} - 2^{54} + 1$$

  • Proved in Lean: Theory/CyclotomicInertia.lean proves p61_prime via kernel-checked Lucas certificates.
  • 2-Adicity: Exactly 54. Supports negacyclic NTT up to $N = 2^{53}$.
  • Fold Congruence: $2^{61} \equiv 2^{54} - 1 \pmod p$, enabling Goldilocks-style reduction with 3 lazy bits. The Revival: By instantiating BFV ciphertexts directly over $p_{61}$, the ciphertext coefficients live in the native proving field! The multilinear PCS (BaseFold over $p_{61}$) binds the ciphertexts directly, closing the $60M Dusk/OtterSec evaluation hole without costly field emulation.

4.4 o7vn Isabelle/MLton Pipeline $\to$ Mechanizes uc-crypthol

uc-crypthol suffered from the semantic transport gap between Isabelle2025-2 and Lean 4. o7vn has perfected an industrial Isabelle $\to$ SML $\to$ MLton pipeline:

  • Code generation from Isabelle theories (src/isabelle/) directly compiles via MLton into standalone, ultra-high-performance native binaries. The Revival: Instead of hand-bridging CryptHOL theorems into Lean Prop fields, we can use o7vn's pipeline to compile CryptHOL-verified cryptographic algorithms directly into native shared libraries that are linked into the node.

4.5 Plonky3-Recursion & Degree-3 Sumcheck

In minidregg/Assurance/AirSumcheckCubic.lean, the degree-3 sumcheck engine unifies GKR fraction-tree layers and Hadamard elementwise zerochecks: $$\text{cubicForm } E\ A\ B\ C\ D = \hat{E} \cdot (\hat{A}\hat{B} + \hat{C}\hat{D})$$ With soundness error $m \cdot 3 / |F|$. This connects directly to ZkmlMatmulSuccinctChecker.lean, which verifies matrix multiplication contractions at $(\mu + \nu)/|F| + \kappa \cdot 3 / |F|$ without reading dense tables.


5. The Revival Roadmap

The table below outlines the precise engineering blueprint for taking back up Breadstuffs' parked subsystems using our modern foundational toolchains:

#SubsystemParked Asset in breadstuffsModern Foundation EnablerTarget Architecture & Revival ActionPriority
R1FHE NTT/PBS Hardwarefhegg-rtl/ (Lean netlist DSL, Golden butterfly/accumulator)obend/ Silicon Elaborator (Obend/Hardware/ReductionMachine.lean)Lower Lean FHE butterfly specs through Core4 $\to$ Leaf.Book $\to$ AutomatonTile grid on eFPGA. Completely bypass handwritten SpinalHDL.P1 (High)
R2vFHE Common-Point Foldfold-opening/ (MLE common point opening, zero sumcheck)zkml-research/ $p_{61}$ field (p61_prime, $127 \cdot 2^{54}+1$)Re-base BFV ciphertexts onto $p_{61}$. Disclose evaluations against a BaseFold PCS over $p_{61}$. Eliminate $\Theta(B \cdot N \cdot L)$ AIR constraints.P1 (High)
R3Post-Quantum Threshold Signaturescrypto-hermine/ & crypto-tanuki/minidregg/ access control & o7vn/ verified relationsImplement an asynchronous network ceremony with weak-VSS to replace the trusted dealer. Deploy Hermine as the threshold validator signature.P2 (Med)
R4Post-Quantum Randomness Beaconcrypto-hashrand/ (Synchronous safety core, HashCR proof)minidregg/ consensus & commit-reveal engineComplete the asynchronous Gather and Approximate Agreement layer. Replace classical BLS beacons with HashRand across the cluster.P2 (Med)
R5CryptHOL UC Verificationuc-crypthol/ (Dregg2_FCom.thy, Isabelle2025-2)o7vn/ Isabelle $\to$ SML $\to$ MLton compiler pipelineCompile CryptHOL-verified cryptographic primitives directly to native code via MLton, eliminating the unmechanized Lean bridge caveat.P3 (Deep)
R6Lightweight Constraint Checkingconstraint-lowering/ (BabyBear borrow chain)minidregg/ Farkas SysCert (Pred/Satisfiable.lean)Replace complex range-check polynomial gadgets with $O(\V\
R7Confidential Clearing Corefhegg-solver/ (PDHG solver, Cert-F certificates)metatheory/Market/ audit & zkml-researchFix the MPC reveal specification split (reconcile argmax vs balance-threshold) and bind Cert-F checking to the deployed DEX clearing loop.P2 (Med)

6. Primary Source Verification & Hash Manifest

All statements and paths in this strand are grounded in the following immutable artifacts:

  1. Breadstuffs Law #1 Enforcement:
  2. Commit: 19439d261d9622f6a08a39df9a02e0bbcc267054 (2026-07-31)
  3. Gate: /Users/ember/dev/breadstuffs/circuit-prove/tests/law1_enforcement_gate.rs
  4. Canonical Doc: /Users/ember/dev/breadstuffs/docs/CIRCUITS-FROM-PROOFS.md
  5. Breadstuffs Core Audit:
  6. Document: /Users/ember/dev/breadstuffs/CORE-AUDIT.md (Workflow wf_68020930-db3, 2026-07-03)
  7. Fixed in: starbridge-v2/src/world/world.rs (LedgerJournal::touched_cells)
  8. Verified Rotated Layout:
  9. Commit: 94217ed3c (2026-07-18)
  10. Source: /Users/ember/dev/breadstuffs/metatheory/Dregg2/Circuit/Emit/EffectVmEmitRotationV3.lean
  11. Isabelle / CryptHOL UC Grounding:
  12. Commit: e17b9da35 (2026-06-07)
  13. Sources: /Users/ember/dev/breadstuffs/uc-crypthol/{Dregg2_FCom.thy, ROOT}
  14. Environment: Isabelle2025-2 + AFP 2025-2
  15. Plonky3 IVC Fork:
  16. Commit: af3f6cc269810d2e3a73baceabb47edb83333578 (2026-06-26)
  17. Source: /Users/ember/dev/plonky3-recursion/README.md
  18. Mini Admission & Farkas Solver:
  19. Memory: ~/.claude/projects/-Users-ember-dev-breadstuffs/memory/project-no-proof-system-in-mini-admission.md
  20. Theorem: /Users/ember/dev/minidregg/Pred/Satisfiable.lean:486 (SysCert.check_sound)
  21. Re-execution: /Users/ember/dev/minidregg/Kernel/Run.lean:316,383 (checkRun_sound, no_accepted_of_output_mismatch)
  22. $p_{61}$ Field Discovery:
  23. Theorem: /Users/ember/dev/minidregg/Theory/CyclotomicInertia.lean (p61_prime, commit 641ceeb)
  24. Spec: /Users/ember/dev/zkml-research/notes/fold-as-opening.md
  25. Silicon Elaborator:
  26. Spec: /Users/ember/dev/obend/SILICON-ELABORATOR.md
  27. Machine: /Users/ember/dev/obend/Obend/Hardware/ReductionMachine.lean
Dossier 17 · Breadstuffs DeOS Surfaces, Starbridge 32-App Ecosystem & Moldable Inspector Diaspora (46 KB)
Working field notes & source index below, with theorem anchors, code pointers, git history, and open questions.

Strand 17 — DeOS Surfaces, Starbridge App Ecosystem, Moldable Inspector, and the Diaspora

Primary-source technical dossier, compiled 2026-11-02 from the repositories, git commits, architecture docs, agent transcripts, and memory corpora across /Users/ember/dev/breadstuffs, /Users/ember/dev/delvetalk, /Users/ember/dev/delvetalk2, /Users/ember/dev/spween, /Users/ember/dev/spw-workbench, /Users/ember/dev/minidregg, /Users/ember/dev/obend, and ~/.claude/projects/-Users-ember-dev-breadstuffs/memory/.


0. Executive Ledger & The Guiding Vision

0.1 Primary Source Ledger

System / SurfaceBreadstuffs Path & Git CommitModern Crate / Diaspora LocationPrimary Invariants & Verification Seams
Starbridge v2 Cockpitstarbridge-v2/ (a7de1c5a4, d5ad5a6e4) (194,109 LOC)obend/crates/obend-gui (2,088 LOC)GPUI ocap master shell, 32 mode tabs, docked dev loop, spotter palette, living card panes
Starbridge App Ecosystemstarbridge-apps/ (1e25f8fc8, 4b80c7095) (114,679 LOC)minidregg/tests/objective-bend-source/32 domain apps; rule: "The answer is never Effect::FooApp"; slot caveats (WriteOnce, Monotonic, AffineLe)
Moldable Inspector & Gadgetsstarbridge-v2/src/presentable.rs (800945db6, ce4aca034)obend-gui/src/viewport.rs7 presentation kinds (RawFields..Source), L1–L10 ladder, IntentDraft -> simulate() -> commit()
Cap-Bounded Reflectiondeos-reflect/ (21ebc84cf, b3ef43204) (1,420 LOC)minidregg DeOS CrownCap-bounded attested read (Inspectable, OcapGraph, Frustum, AffordanceSurface); no omniscient leaks
Renderer-Agnostic View-Treedeos-view/ (ecbbe2252, c5ab6469e) (19,963 LOC)obend-gui/src/server.rsDual-renderer (native GPUI vs web HTML) over serializable ViewNode tree; progressive disclosure
Zed Full Workspace Embeddeos-zed-full/ (3190aa849, 302aea28c) (4,774 LOC)Parked / Target for spw-workbenchReal Zed Workspace over FirmamentZedFs; files = cells, file save = verified turn
Matrix Rehydratable Membranedeos-matrix/ (c330befc5, afcac822e) (7,706 LOC)delvetalk2 AT Proto JournalChat rooms = cells, messages = turns; messages carry rehydratable world-fork membranes (shared_fork.rs)
DeOS Homeserver Graindeos-homeserver/ (07942f63c, f1c2a5ffe) (13,484 LOC)Sandstorm-style sovereign grainMatrix homeserver as an isolated Sandstorm grain hosted on sovereign cell storage
Software-Rasterized Webviewservo-render/ (83179110f, 5ae26d4ee) (24,749 LOC)ParkedSWGL (swgl 0.68.0) CPU rasterizer; renders RGBA8 frames into CompositorPd digest; no GPU/EGL
Android Cell (Graphideos)android-cell/ (6fae061d0, 8cd45ddee) (11,066 LOC)ParkedAndroid APKs as cap-confined cells; AppFactory, PermGate cap badges, SystemUI cap-chrome
Sandstorm Grain Integrationsandstorm-bridge/, sandstorm-package/ (028bc0271)Extracted for Mini hosting (Ember 09-26)Real .spk parser (xz, Ed25519, Cap'n Proto); grain /var = cell umem heap; Powerbox = attenuating caps
Verifiable Narrative DSLspween-dregg/ (4ddb4f1f6, dad9fd8db) (8,183 LOC)~/dev/spween, delvetalk2 scenesInteractive fiction DSL on world-cells; choices = verified turns; un-retconnable receipt chains
DrEX Terminal v2drex-web-v2/ (da95a0fa7, 6f04af76f) (1,790 LOC)Parked / Target for obend-guiPreact + Signals DEX terminal; Open/Shielded/Dark tier-dial; window.dregg extension handshake
Fair Launch Engine Portallaunchpad-web/ (e1247989b, 8e1850596) (2,245 LOC)ParkedUniform-price fair-launch indexer driven by live node turn streams (POST /turn/submit)
Light Client Portalportal/ (1f39d35f2, df7118b88) (12,006 LOC)ParkedIn-tab recursive-STARK light client (portal.dregg.studio) evaluating cryptographic receipts
Browser Wallet & Brokerextension/ (06b366e54, 7593b4c1d) (74,905 LOC)ParkedWebExtension (window.dregg); BIP-39 mnemonic identity, signing, and capability brokerage
Town of Inhabited ObjectsN/A (Modern Synthesis)delvetalk2 (branch foundation)Post-cockpit subtraction: text & cards, spells, receipts, Hob the reader, AT Protocol repository

0.2 The Founding Philosophy: 1999 AOL Wonder × Pharo Smalltalk Liveness

In ~/.claude/projects/-Users-ember-dev-breadstuffs/memory/project-deos-ux-vision.md (origin session 3b829648-9d41-4dad-bd2e-1a3eb7f618ab, recorded 2026-06-19), Ember articulated the existential purpose and user experience of DeOS:

"i am the person who needs this substrate. i'll be using this os for a while." *The "who needs this / killer app" question is ANSWERED and it dissolves the biggest strategic lack: ember IS the user. deos is her daily-driver sovereign OS — she will live in it. Not a fallback, the Engelbart/Pharo/Houyhnhnm THESIS: you build the tools you live in; sovereign computing is for the sovereign, and the sovereign is here.* *The feeling — 1999 AOL as a 4-5-year-old. "WOW!! THIS IS AMAZING!! i had no comprehension. just clicking around. absorbing. not understanding why sometimes the back button stopped working (a new window was opening over it!)." Running RedHat off his dad's machine. Browser, website, flash. The deos UX is supposed to feel like THAT — the user "has no idea wtf is going on" and it's wonderful. Wonder precedes comprehension; you click, you absorb, the system rewards exploration. The deos objects you click around: badges, powerboxes, living screenshots (the rehydratable frustum-snapshots), cells, affordances.* *The substrate — Pharo/Smalltalk LIVENESS. "the smalltalk experience. it's supposed to be pharo-ish." Pharo = a live, image-based, self-describing environment: every object is a live, inspectable, modifiable thing; you can halt anywhere, inspect, change, continue; the whole system is malleable from within itself. deos is ALSO an IDE, debugger, etc. — exactly as Pharo's environment IS its own live IDE/inspector/debugger.*

This dual thesis governed every line of DeOS UI code written between May and August 2026: a user experience combining the uninhibited, tactile magic of 1990s desktop computing with the mathematical rigor of an object-capability operating system where every button press is a verified state transition (a turn), every UI panel is an inspectable cell, and every window can be serialized into a cryptographic snapshot and sent over chat to be rehydrated and executed on another machine.


1. What Lived in Breadstuffs: The 31 Environments, Crates & Surfaces

Breadstuffs was not merely a cryptographic kernel; it was a complete, sovereign operating system environment comprising over 550,000 lines of UI, environment, application, and tooling code. This ecosystem spanned six distinct functional quadrants.

text
+---------------------------------------------------------------------------------------------------+
|                                  THE BREADSTUFFS DeOS CONTINUUM                                   |
+---------------------------------------------------------------------------------------------------+
|  [Desktop & Execution Cockpit]          |  [Cap-Bounded Reflection & View]                        |
|  - starbridge-v2 (194k LOC, GPUI shell) |  - deos-reflect (1.4k LOC, pure AST/Ocap projection)    |
|  - servo-render (24k LOC, SWGL frame)   |  - deos-view (19.9k LOC, native GPUI + web HTML)        |
|  - android-cell (11k LOC, Graphideos)   |  - deos-js / runtime (21.5k LOC, SpiderMonkey applets)  |
|  - dregg-tui (1.1k LOC, Ratatui CLI)    |  - deos-web-cells (4.6k LOC, live DOM cells)            |
+-----------------------------------------+---------------------------------------------------------+
|  [Developer Panes & Inhabitation]       |  [Multiplayer Communications & Grains]                  |
|  - deos-zed & zed-full (11.4k LOC, FS)  |  - deos-matrix (7.7k LOC, rehydratable membranes)       |
|  - deos-terminal (3.3k LOC, Alacritty)  |  - deos-homeserver (13.4k LOC, grain-hosted Matrix)     |
|  - deos-hermes (24k LOC, agent harness) |  - sandstorm-bridge/package/serve (10.3k LOC, SPKs)     |
+-----------------------------------------+---------------------------------------------------------+
|  [Starbridge 32-App Ecosystem]          |  [Web Portals, Wallets & Financial Frontends]           |
|  - starbridge-apps (114.6k LOC)         |  - extension (74.9k LOC, browser wallet & broker)       |
|  - app-framework (21.8k LOC, DSI)       |  - portal (12k LOC, in-tab STARK light client)          |
|  - starbridge-web-surface (17k LOC)     |  - drex-web & drex-web-v2 (5k LOC, DEX terminals)       |
|  - spween-dregg (8.1k LOC, CYOA engine) |  - launchpad-web (2.2k LOC, fair-launch turn stream)    |
+---------------------------------------------------------------------------------------------------+

1.1 The Native Desktop Cockpit & Rendering Engine

1. starbridge-v2 (/Users/ember/dev/breadstuffs/starbridge-v2, 194,109 LOC)

  • Scaffold Commit: a7de1c5a4 (2026-06-13); Last Touches: d5ad5a6e4 (2026-09-30).
  • Core Architecture: The flagship native desktop application built using Zed's GPUI framework. It housed 115 source files including the master cockpit (src/cockpit/), containing 21 modules (panels_workspace.rs, panels_main.rs, panels_moldable.rs, panels_web.rs, panels_devtools.rs, panels_app_launcher.rs, live.rs, construct.rs, render.rs, shell_ops.rs).
  • DeOS Desktop Subsystem (src/deos_desktop/, 29 modules): Implemented the full Pharo/Smalltalk-style desktop:
  • world_explorer.rs: Visual hierarchy walker over sovereign cells, authority delegations, and receipts.
  • spotter.rs: Universal command palette and search indexing every Presentable object.
  • halo.rs: Direct-manipulation ring surrounding on-screen cells (tear off, inspect, fork, actuate).
  • systemui_chrome_render.rs & android_window.rs: Cap-badge status bars and quick-setting pull-downs.
  • matrix_room.rs, agent_room.rs, exchange_floor.rs, docgraph_view.rs, face_scroll.rs, toasts.rs.
  • Inhabitation & Rehydration: src/shared_fork.rs and src/distributed_card.rs enabled live cockpit tabs to be detached into independent OS windows (tear_off_tab_unleased), snapshotted to disk, or transmitted across machines.
  • The seL4 Framebuffer Keystone: Commit d92d0526c (2026-06-19) closed the #1 precious milestone: wiring GPUI's offscreen software render directly into the seL4 QEMU ramfb framebuffer (deos-live.system), allowing the full Starbridge v2 cockpit to boot on a formally verified microkernel without an X11 or Wayland host.

2. servo-render (/Users/ember/dev/breadstuffs/servo-render, 24,749 LOC)

  • Origin Commit: 83179110f (2026-06-14); Key Hardening: 5ae26d4ee (2026-07-27).
  • Architecture: A hardware-independent web rendering pipeline designed to run inside an unprivileged seL4 Protection Domain (PD).
  • The SWGL Core: Used Mozilla WebRender's C++17 software rasterizer (swgl 0.68.0) in swgl_context.rs. It implemented gleam::gl::Gl, compiling pure C++ into a CPU-only rasterizer that rendered HTML/CSS directly into a caller-owned RGBA8 buffer (Vec<u8>).
  • Attested Output: In compositor_seam.rs:
rust
  pub fn present_frame(frame: &RgbaFrame, region: Rect) -> FrameReceipt {
      let digest = blake3::hash(&frame.pixels);
      // Digest committed to seL4 CompositorPd gate (T1 non-overlap, T2 label, T3 focus)
      FrameReceipt { digest, bounds: region, timestamp: now() }
  }
  • Libservo Elephant: Contained an opt-in feature --features libservo in webview.rs wiring into SpiderMonkey (mozjs) and Servo's layout engine. Because of mozjs compilation overhead (multi-GB C++ build), SWGL was cleanly decoupled into swgl-standalone so pixel rendering could be tested instantly without building SpiderMonkey.

3. android-cell (Graphideos) (/Users/ember/dev/breadstuffs/android-cell, 11,066 LOC)

  • Origin Commit: 6fae061d0 (2026-06-24); Memory: project-graphideos-android-epoch.md.
  • The Graphideos Concept: Running unmodified Android applications within cap-confined cell sandboxes. servo : web :: android-runtime : android.
  • Key Modules:
  • appfactory.rs: Minted an android-cell from an APK's manifest descriptor; declared permissions defined the exact initial capability set (CapBadgeSet), eliminating ambient Android authority.
  • permgate.rs: Interposed over Android's Context.checkSelfPermission. If a cap badge was dim, the runtime returned PERMISSION_DENIED in-process. Handing over a permission minted a genuine Effect::GrantCapability through the kernel TurnExecutor.
  • intentgate.rs & broadcastgate.rs: Intercepted Android Intents and Binder RPCs, routing them through verified capability checks.
  • runtime/: Shipped MacOsEmulatorRuntime (driving Google Android Emulator over Hypervisor.framework) and CapturedFrameRuntime (headless PNG/RGBA8 tile pipe).

4. dregg-tui (/Users/ember/dev/breadstuffs/dregg-tui, 1,132 LOC)

  • Ratatui-based terminal cockpit displaying the active cell graph, ledger height, witness cursor, and recent receipts for headless server environments and remote SSH sessions.

1.2 Cap-Bounded Reflection & The View-Tree Substrate

5. deos-reflect (/Users/ember/dev/breadstuffs/deos-reflect, 1,420 LOC)

  • Origin Commit: 21ebc84cf (2026-06-23); Fix: b3ef43204 (2026-08-01).
  • Purpose: Cap-bounded, attested reflection over cell substance, completely independent of GPUI. Crawling was strictly per-viewer: a principal can only observe what their authority reaches.
  • Core Types (src/lib.rs):
rust
  pub use substance::{Field, FieldValue, Inspectable, ObjectKind, reflect_cell};
  pub use graph::{GraphEdge, GraphLayer, GraphNode, OcapGraph};
  pub use frustum::Frustum;
  pub use affordances::{Affordance, AffordanceSurface, EffectSummary};
  pub use present::{Presentation, PresentationBody, PresentationKind, ReflectedCell};
  • The Attested Redaction Principle: When inspecting private or committed fields, deos-reflect emits the cryptographic commitment (Poseidon2 / Blake3), never the cleartext value.

6. deos-view (/Users/ember/dev/breadstuffs/deos-view, 19,963 LOC)

  • Origin Commit: ecbbe2252 (2026-06-23); Last Touches: c5ab6469e (2026-08-06).
  • Core Thesis: The card is renderer-independent. deos-js produces declarative UI trees (tree::ViewNode), and deos-view renders that data into pixels or markup.
  • Dual Renders:
  • native (GPUI): bridge::build_live_view executes SpiderMonkey JS applets, extracts ViewNode, and render::AppletView translates nodes (vstack, button, text, bind) into GPUI widgets. Buttons bind directly to World::commit_turn.
  • web (HTML/DOM): web::render_card_document translates the exact same ViewNode tree into a standalone .html string with embedded styling and SVG badges.
  • Progressive Disclosure: ViewNode::Adept, Disclosure::Simple | Disclosure::Adept, and the disclose() AST transform allowed UI elements to hide developer/cryptographic details from casual users while allowing power users to expand raw state constraints and Lean invariants.

7. deos-js, deos-js-core, deos-js-runtime (21,526 LOC combined)

  • Embedded SpiderMonkey runtime executing unprivileged userspace applets.
  • Exposed the deos.ui.* DOM building API, deos.turn.* intent builder, and deos.cell.* query layer.
  • src/attach.rs: Enabled an autonomous agent (Claude Sonnet 4.5 via ACP/MCP) to attach to a live Rc<RefCell<World>>, inspect cells through deos-reflect, synthesize a JavaScript applet, execute it, and commit verified state transitions (0d03f6ee...).

1.3 Developer Panes, Confined Inhabitation & Editors

8. deos-zed (6,672 LOC) & deos-zed-full (4,774 LOC)

  • Origin Commit: 3190aa849 (2026-06-23); Test Harness: 302aea28c (2026-09-30).
  • The Vision: Embedding the full, real Zed editor stack directly into the DeOS cockpit.
  • FirmamentZedFs (src/firmament_zed_fs.rs): Implemented Zed's asynchronous fs::Fs trait over sovereign cells:
  • Files and directories mapped to cells in the sovereign graph.
  • Buffer saves invoked verified kernel turns:
rust
    impl fs::Fs for FirmamentZedFs {
        async fn save(&self, path: &Path, content: &Rope, ...) -> Result<()> {
            let patch = dregg_doc::diff(&self.read(path).await?, content);
            let turn = TurnBuilder::new().apply_patch(self.cell_for_path(path)?, patch).build();
            self.executor.commit_turn(turn).await?;
            Ok(())
        }
    }
  • Validated by three headless integration tests (cargo test --features full-zed): tests/firmament_zed_fs.rs, tests/project_over_cells.rs, and tests/full_workspace_over_cells.rs. The full Zed Workspace, project_panel, outline_panel, and terminal_view mounted cleanly over cells-as-files.

9. deos-terminal (/Users/ember/dev/breadstuffs/deos-terminal, 3,309 LOC)

  • Integrated Alacritty PTY terminal dock pane. Allowed the developer to pop open a shell with ⌘K, executing local commands inside a cap-bounded sub-environment.

10. deos-hermes (/Users/ember/dev/breadstuffs/deos-hermes, 24,012 LOC)

  • The confined agent host. Hermes ran autonomous LLM reasoning loops inside an isolated Protection Domain. All external actuation (filesystem, network, editor) was mediated via capability handles, with ambient authority denied (verdict 0xf). Every agent action emitted an irrevocable, verifiable turn receipt.

1.4 Multiplayer Communications & Sovereign Grains

11. deos-matrix (7,706 LOC) & deos-homeserver (13,484 LOC)

  • Origin Commit: c330befc5 (2026-06-22); Memory: project-deos-self-hosting-epoch.md.
  • The Concept: Native Matrix chat (matrix-rust-sdk) re-pilled around sovereign cells:
  • Matrix Rooms = Sovereign Cells; Matrix Identities = Cell Cap Badges; Messages = Verified Turns.
  • The Rehydratable Membrane (src/membrane.rs): The crowning multiplayer primitive. A chat message could encapsulate a membrane: a frustum-culled, cap-bounded snapshot of a live world-fork. The recipient received the message, clicked the membrane, and the cockpit rehydrated the world-fork in-memory. The recipient executed turns within the fork, and a subsequent stitch operation merged the divergent branches back into the sender's ledger using categorical pushouts.
  • deos-homeserver: A Matrix homeserver designed to run as an unprivileged, persistent Sandstorm-style grain on top of the cell store (docs/deos/GRAIN-HOMESERVER.md), making communications self-hosted without relying on third-party Matrix homeservers.

12. Sandstorm Integration: sandstorm-bridge, sandstorm-package, sandstorm-serve (10,393 LOC combined)

  • Origin: Began in May 2026; standalone package parser extracted by Ember in commit 028bc0271 (2026-09-26: "Extract reusable signed Sandstorm package parser for Mini hosting").
  • Mechanism:
  • sandstorm-package: Decompressed real .spk packages (8-byte magic Spk1, xz decompression, Ed25519 signature verification where App ID = public key, and Cap'n Proto archive decoding).
  • sandstorm-bridge: Mapped Sandstorm grain concepts directly to DeOS cells:
  • Grain /var directory = Sovereign cell umem heap.
  • Grain lifecycle (create, wake, sleep, delete) = Cell lease and persistence management.
  • Sandstorm Powerbox = Attenuable credentials (dga1_..., ed25519 caveat chains).
  • HTTP Bridge: Interposed X-Sandstorm-* identity and permission headers derived from the holder's cell capability.

1.5 The Starbridge App Ecosystem & Web Surfaces

13. app-framework (21,851 LOC) & starbridge-apps (114,679 LOC)

  • Origin Commit: 1e25f8fc8 (2026-05-24); Census: docs/deos/STARBRIDGE-APPS-CENSUS.md.
  • The Immutable Law:

"The answer is never Effect::FooApp." *When an app wants a domain Effect, the missing primitive is the generic one (Caveat, StateConstraint, Authorization, Factory) it would compose from. Every starbridge-app must be buildable from dregg-native primitives only.*

  • The 32 Apps:
  • nameservice: Hierarchical domain registration with WriteOnce and Monotonic slot caveats.
  • identity: Selective credential disclosure, verifiable presentation, and revocation.
  • subscription: Cap-metered bounded ring buffer for pub/sub message consumption.
  • governed-namespace: Threshold multisig voting on namespace registration.
  • compartment-workflow-mandate: Clearance-graph workflow progression and spend policies.
  • storage-gateway-mandate: Bounded volume read/write gateway for blob storage.
  • privacy-voting: Monotonic poll counters with per-voter one-vote-per-cell caveats.
  • bounty-board: First-claimer-wins bounty lifecycle (POSTED -> CLAIMED -> SUBMITTED -> PAID).
  • compute-exchange: Escrow budget state machine (AffineEq(PAID + REFUNDED == BUDGET)).
  • gallery: Commit-reveal art curation using sealed submission boards.
  • sealed-auction: Sealed-bid uniform price auction clearing.
  • domains: Distributed DNS-style authoritative zone cells.
  • edge-mandate: Attenuated execution mandates at network edge nodes.
  • escrow-market: Multi-party conditional asset escrow.
  • first-room: Smalltalk-style beginner world and cell tutorial.
  • guard: Capability revocation gates and watchdog contracts.
  • kvstore: Verified content-addressed key-value store.
  • org: DAO structure with role hierarchies and treasury allocations.
  • polis: Quadratic voting and Deliberative polling.
  • site-host: Static website hosting served directly from cell heaps.
  • supply-chain-provenance: Multi-hop physical asset custody tracking.
  • swarm-orchestration: Multi-agent work distribution and receipt settlement.
  • tool-access-delegation: Fine-grained attenuation of CLI/API tool execution.
  • tussle: Competitive multi-agent game state machine.
  • execution-lease: Computron resource metering and lease renewals.
  • vat: E-style isolated event-loop containers with CapTP messaging.
  • agent-orchestration: Hierarchical subagent scheduling.
  • agent-provenance: Verification of LLM model signatures and prompts.
  • billing: Usage-based micro-invoicing.
  • branch-stitch-multiplayer: Multi-user document branch merge demonstrator.
  • shared: Common Preact inspectors and JSON factory descriptors.
  • starbridge-web-surface (17,030 LOC): Browser-based dregg:// cell URI resolver and transclusion.

14. deos-leptos (4,108 LOC) & deos-web-cells (4,667 LOC)

  • deos-leptos: Explored Rust web frontend reactivity: Cell state = Leptos RwSignal; Affordances = reactive UI components; Rehydration = per-viewer SSR hydration.
  • deos-web-cells: Implemented live DOM snapshot cells (BundleKind::LiveDomSnapshot), enabling declarative HTML/DOM subtrees to be content-addressed, transcluded across documents, and stored directly on the cell ledger.

15. Frontends & Portals: drex-web-v2, launchpad-web, portal, extension

  • drex-web-v2 (1,790 LOC, da95a0fa7): Clean Preact + @preact/signals trading interface. Implemented a 3-tier view dial (Open = public order book, Shielded = blurred liquidity amounts, Dark = sealed-bid commit-reveal). Handshook with window.dregg.
  • launchpad-web (2,245 LOC, e1247989b): Fair-launch token interface driven directly by live node turn streams (POST /turn/submit). Bypassed traditional EVM bonding curves in favor of batch-cleared uniform-price auctions.
  • portal (12,006 LOC, 1f39d35f2): Trustless network explorer (portal.dregg.studio) embedding the dregg-wasm recursive STARK light client. Verified block headers and turn receipts locally in the tab.
  • extension (74,905 LOC, 06b366e54): Firefox/Chrome browser extension exposing window.dregg. Managed BIP-39 mnemonic phrases, signed transactions, and brokered capability delegations.
  • spween-dregg (8,183 LOC, 4ddb4f1f6): Bound the Spween narrative-choice DSL to sovereign world-cells. Every player dialogue choice was executed as a verified turn, producing an unforgeable, un-retconnable receipt chain. Included VoteEngine for collective Twitch-plays-style storytelling.

2. The Moldable Inspector & Gadget Framework

In starbridge-v2/src/presentable.rs (commit 800945db6) and docs/deos/INSPECTOR-FRAMEWORK.md, Ember adapted Tudor Gîrba's Glamorous Toolkit (GT) and Pharo moldable inspector philosophy to the cryptographic ledger.

text
+---------------------------------------------------------------------------------------------------+
|                                  THE MOLDABLE INSPECTOR PIPELINE                                  |
+---------------------------------------------------------------------------------------------------+
|                                                                                                   |
|   +-------------------------------------------------------------------------------------------+   |
|   |                      The 7 Presentation Kinds (enum PresentationKind)                     |   |
|   |  1. RawFields: Mandatory structural floor (field-tree with attested redactions)           |   |
|   |  2. Graph: Ocap delegations, cell reachability closure, state machine transitions         |   |
|   |  3. DomainVisual: High-level domain visualizer (ballot box, token ring, clearance tape)   |   |
|   |  4. Affordances: Cap-gated message roster; illuminated = allowed, dimmed = refused       |   |
|   |  5. Provenance: Irrevocable receipt chain, turn author signatures, blame graph            |   |
|   |  6. Invariant: Active state constraints, Lean theorem checks, conservation proofs        |   |
|   |  7. Source: Underlying Lean / Objective Bend / AST source definition                      |   |
|   +-------------------------------------------------------------------------------------------+   |
|                                                |                                                  |
|                                                v                                                  |
|   +-------------------------------------------------------------------------------------------+   |
|   |                         Interactive Gadgets (trait CommittingGadget)                      |   |
|   |                                                                                           |   |
|   |    [User Input / Dial]  --->  IntentDraft  --->  simulate()  --->  commit_turn()          |   |
|   |                                                      |                    |               |   |
|   |                                             (Predict Turn)         (Verified Turn)        |   |
|   +-------------------------------------------------------------------------------------------+   |
|                                                |                                                  |
|                                                v                                                  |
|   +-------------------------------------------------------------------------------------------+   |
|   |                         The L1--L10 Shipped Inspector Ladder                              |   |
|   |   L1: presentable.rs         L2: predicate_composer.rs      L3: turn_builder.rs           |   |
|   |   L4: cap_inspector.rs       L5: cell_inspector.rs          L6: receipts_inspector.rs     |   |
|   |   L7: token_inspector.rs     L8: federation_inspector.rs    L9: circuit_inspector.rs      |   |
|   |                              L10: settlement_inspector.rs                                 |   |
|   +-------------------------------------------------------------------------------------------+   |
|                                                                                                   |
+---------------------------------------------------------------------------------------------------+

2.1 The Core Traits: Presentable, Gadget, CommittingGadget

Unlike standard debuggers that provide a static, one-size-fits-all JSON or property-grid tree, every DeOS protocol object implemented Presentable:

rust
// starbridge-v2/src/presentable.rs
pub trait Presentable {
    /// Returns the set of named presentation faces this object supports.
    fn presentations(&self, ctx: &PresentationContext) -> Vec<Presentation>;

    /// Render a specific presentation face.
    fn render_presentation(&self, kind: PresentationKind, ctx: &PresentationContext) -> PresentationBody;
}

pub trait Gadget {
    type Output;
    fn render_editor(&mut self, ui: &mut GadgetUi) -> Option<Self::Output>;
}

pub trait CommittingGadget: Gadget {
    /// Predict-then-commit: every UI mutation stages an IntentDraft,
    /// verifies the turn against local invariants, and commits to the ledger.
    fn stage_intent(&self) -> Result<IntentDraft, GadgetError>;
    fn commit(&mut self, executor: &mut TurnExecutor) -> Result<Receipt, GadgetError>;
}

2.2 The Shipped L1–L10 Ladder

The framework successfully replaced monolithic UI views with modular inspector modules:

  • L1 — Presentation Spine (presentable.rs): Base registry, presentation kinds, and Spotter palette.
  • L2 — Predicate Composer (predicate_composer.rs): Visual builder for StateConstraint predicates (WriteOnce, Monotonic, AffineLe).
  • L3 — Turn Builder (turn_builder.rs): Effect stage and intent drafter.
  • L4 — Cap Inspector (cap_inspector.rs): Visualized capability attenuation chains and caveats.
  • L5 — Cell Inspector (cell_inspector.rs): Live cell slot viewer and value inspector.
  • L6 — Receipts Inspector (receipts_inspector.rs): Turn receipt auditor and execution trace walker.
  • L7 — Token Inspector (token_inspector.rs): Balance rings and conservation law verifier.
  • L8 — Federation Inspector (federation_inspector.rs): Blocklace DAG consensus viewer.
  • L9 — Circuit Inspector (circuit_inspector.rs): AIR constraint polynomial viewer.
  • L10 — Settlement Inspector (settlement_inspector.rs): Cross-chain bridge escrow and settlement status.

3. The Diaspora: How Ideas & Surfaces Spread Into Modern Projects

When the monolithic Breadstuffs repository froze in August–October 2026, its ideas did not die; they dispersed into specialized downstream projects.

text
                                    /Users/ember/dev/breadstuffs
                                                 |
         +--------------------+------------------+-------------------+--------------------+
         |                    |                                      |                    |
         v                    v                                      v                    v
  ~/dev/delvetalk2      ~/dev/spween & spw-workbench           ~/dev/minidregg     ~/dev/obend/obend-gui
  ----------------      ----------------------------           ---------------     ---------------------
  - Inhabited Town      - Standalone DSL crate                 - "DeOS Crown"      - Lightweight runner
  - Plain-words Law     - VS Code extension & LSP                Lean theorems     - 10 ProjectionKinds
  - Cards & Spells      - Semantic brace operators             - Compressed core   - LawDial controls
  - AT Protocol repo    - Narrative state trees                  (<150k lines)     - SVG / HTML views
  - "No coins" rule     - Collective branch choice             - Zero twins        - Picard-Lindelof ODE

3.1 Diaspora Destinations

1. delvetalk & delvetalk2 (/Users/ember/dev/delvetalk2, branch foundation)

  • The Inhabited World Reborn: DelveTalk inherited the soul of DeOS—the world of durable, programmable, inhabited objects. But it radically subtracted the heavy GUI scaffolding:
  • Monolithic GPUI replaced by Plain Text & Cards: Objects are described by cards; interactions take the form of natural-language replies or concise "spells" (delvetalk garden-1 plant).
  • Hob the Reader: Ingestion layer translating natural language into admitted action spells.
  • AT Protocol Repository: Town state federated as signed AT Protocol records (did:web:gsb.fg-goose.online).
  • "No coins" Rule: Strict policy against exposing raw hashes, DIDs, or blockchain jargon on user-facing surfaces.

2. spween & spw-workbench (/Users/ember/dev/spween, /Users/ember/dev/spw-workbench)

  • spween: Decoupled into an elegant, educational DSL parser and interpreter for interactive fiction, maintaining the trait-based EffectHandler abstraction.
  • spw-workbench: Evolved into a dedicated organelle for .spw workspaces, featuring a complete Language Server Protocol (LSP) server, CLI, runtime, and VS Code extension (extensions/vscode-spw).

3. minidregg (/Users/ember/dev/minidregg)

  • The DeOS Crown Preserved in Lean: In minidregg/ATLAS.md (§3, item 14), Ember codified what survived Breadstuffs:

*"The deos crown: surface-as-cap, membrane non-amplification, replayedDeterministic_iff_confined (confinement IS the deterministic-replay fragment), affordance-soundness; DocMerge as LUB / pushout products; fog-of-war non-interference. Identity: affordance = cap-gated effect-template; firing = turn; turn = receipt — one machinery, three faces (human/authored/agentic)."*

  • tests/objective-bend-source/DeosCounter.obend: Directly ported Breadstuffs' deos-js counter applet into formal Objective Bend syntax, modeling turns as Activity<Plan, Response, Nat>.

4. obend-gui (/Users/ember/dev/obend/crates/obend-gui, 2,088 LOC)

  • The Modern Viewport Successor: Replaced starbridge-v2's massive 194k-line GPUI monolith with a tight, modular notebook runner and hardware CAD projection previewer.
  • Supports 10 distinct ProjectionKinds: Circuit, Netlist, Timing, Gdsii, Waveform, CellQuote, NotebookCell, ProofTree, ContractBadge, ClearingLedger.
  • Implements interactive SVG/HTML rendering for Picard–Lindelöf ODE flowpipe slices, 5-axis CNC waypoints, eFPGA heatmaps, and live .free / .press / .locked LawDial controls.

3.2 Lessons Learned in the Transition

1. Why Rigid Monolithic Shells Failed

  • The Build Lock Nightmare: In July 2026, local developer iteration ground to a halt because compiling starbridge-v2 locked the entire target directory. Building the desktop cockpit pulled GPUI, Alacritty, Servo, SpiderMonkey, Matrix SDK, and the Lean kernel archive (libdregg_lean.a), resulting in a 292MB binary that took minutes to link and frequently jammed (reference-persvati-remote-build).
  • Duplicate Dependency Crises: Direct dependencies collided with nested dependencies (e.g. workspace GPUI adb902652 vs gpui-component's internal 54fbcb6943), requiring complex workspace overrides.
  • Immediate-Mode UI Re-execution Overhead: Commit f27fd4169 documented that the cockpit repainted carded surfaces as frozen snapshots every frame, BLAKE3-hashing the entire state root on every render tick because projection memoization had not been integrated.
  • The Autonomous Swarm Failure: In ~/.claude/projects/-Users-ember-dev-delvetalk/memory/ember-working-style.md:21:

*"Three hours of autonomous codex produced 45 commits and a 124K-line repo ember called 'horseshit' and 'chug'; the failure was orchestration and refusal to subtract, not effort."*

The remedy was radical subtraction: separate the data models, drop monolithic GUI frameworks, and deliver user agency through lightweight, composable surfaces.

2. Why GPUI Cockpit Builds Needed Discipline

  • Monolithic cockpits must not bundle full editor workspaces, full web browsers, and full chat servers into the primary binary. The split pioneered by deos-view (declarative ViewNode tree) and obend-gui (HTTP/WebSocket server streaming interactive SVG/HTML viewports) demonstrated that rendering belongs at the edge, while state remains pure, verifiable data.

3. How Sandstorm .spk Grains Evolved

  • Sandstorm proved the elegance of capability-secured application grains with private /var heaps. However, Sandstorm grains lacked decentralized, light-client witnessability—users had to trust the central Sandstorm supervisor.
  • In DeOS, grains became content-addressed sovereign cells with verifiable umem heaps. In minidregg and delvetalk2, this evolved into sovereign objects where state transitions are mathematically checked against a plain-words law and recorded on an AT Protocol or blocklace journal.

4. What Got Dropped, Parked, or Underelaborated

During the late-summer 2026 pivot to kernel verification and Mini/Nano rebuilds, several advanced surfaces were parked. These assets represent hundreds of thousands of lines of high-value work:

  1. deos-zed & deos-zed-full (11,446 LOC): The proven integration of Zed's Workspace over FirmamentZedFs. Parked due to GPUI/Zed upstream churn. Left behind: a working editor where saving a file emits a verified Pijul/DocCore patch turn.
  2. deos-matrix & deos-homeserver (21,190 LOC): The rehydratable membrane chat client and grain-hosted Matrix homeserver. Parked when the consensus layer detached from the legacy node. Left behind: the UI mechanism for sending live, forkable world branches inside chat messages.
  3. deos-reflect (1,420 LOC): The pure, cap-bounded reflection engine. Parked during the crate-split campaign. Left behind: the only audit-proven inspector that redacts secrets and graphs capability blast radiuses without pulling GPUI.
  4. servo-render (24,749 LOC): Software-rasterized web browsing using WebRender SWGL. Parked due to SpiderMonkey build complexity. Left behind: working C++17 software rasterization emitting RGBA8 tiles verified by cryptographic digests.
  5. android-cell (Graphideos, 11,066 LOC): Android container/emulator isolation with capability badging. Parked as out-of-scope for the kernel bakeoff. Left behind: real Context.checkSelfPermission interposition and SystemUI capability status bars.
  6. The 34 Apps in starbridge-apps (114,679 LOC): 32 complete domain applications (voting, auctions, escrow, identity, bounty boards, compute markets) adhering to "The answer is never Effect::FooApp". Left behind: complete factory descriptors and slot caveats (WriteOnce, Monotonic, AffineLe).
  7. drex-web-v2 & launchpad-web (4,035 LOC): Modern Preact + Signals trading terminals with 3-tier dials (Open/Shielded/Dark) and live turn-stream fair launches.
  8. extension & portal (86,911 LOC): Browser extension capability broker and in-browser recursive STARK light client verifier.
  9. moldable-inspector (L1–L10 ladder): The 7 presentation faces and gadget composition spine, currently locked inside the legacy starbridge-v2 crate.

5. Concrete Revival Roadmap: Re-Attaching Surfaces to Obend, Mini & DelveTalk2

The diaspora surfaces can be systematically revived by mounting them onto modern hosts: obend-gui, delvetalk2, minidregg, and spw-workbench.

5.1 The Master Revival Matrix

Parked Crate / AssetBreadstuffs LOC & Key EntryModern Target HostNew Adapter Seam / MechanismRevival Tier & Milestone
deos-reflect1,420 LOC<br>deos-reflect/src/lib.rsminidregg & delvetalk2Bind Inspectable & Frustum over DelveTalk WorldObjects and Lean ASTsTier 1 (Immediate)<br>Milestone: Cap-bounded reflection over DelveTalk cards
starbridge-apps (8 Core Seed Apps)~35,000 LOC<br>starbridge-apps/{nameservice, identity, bounty-board, privacy-voting}delvetalk2 & minidreggCompile FactoryDescriptors into Objective Bend .obend record/law definitionsTier 1 (Immediate)<br>Milestone: Bounty board & voting running as DelveTalk objects
moldable-inspector (Spine & Presentable)~8,000 LOC<br>starbridge-v2/src/presentable.rsobend/crates/obend-guiMap the 7 PresentationKinds to obend_gui::ProjectionKind and SVG/HTML viewsTier 1 (Immediate)<br>Milestone: Multi-face inspection of Obend notebooks & Lean proofs
spween-dregg8,183 LOC<br>spween-dregg/src/lib.rsdelvetalk2 & ~/dev/spweenRe-wire spween::EffectHandler to DelveTalk Activity<Plan, Response>Tier 1 (Immediate)<br>Milestone: Verifiable CYOA scenes embedded in Delve.town cards
deos-view (Web/HTML Render Path)~9,000 LOC<br>deos-view/src/web.rsobend-gui/src/server.rsTranslate ViewNode trees into obend-gui interactive HTML/SVG viewportsTier 2 (Near-Term)<br>Milestone: Declarative card rendering in Obend browser interface
sandstorm-package2,248 LOC<br>sandstorm-package/src/lib.rsminidregg hostingReuse Ember's clean Sep 26 standalone parser (028bc0271) to unpack .spk bundlesTier 2 (Near-Term)<br>Milestone: Loading signed legacy Sandstorm tools into cell heaps
drex-web-v2 & launchpad-web4,035 LOC<br>drex-web-v2/src/app.jsobend-gui web clientReplace legacy RPC with obend-gui WebSocket /ws/eval and DelveTalk feedsTier 2 (Near-Term)<br>Milestone: 3-tier dial (Open/Shielded/Dark) trading terminal
deos-matrix Rehydratable Membrane7,706 LOC<br>deos-matrix/src/membrane.rsdelvetalk2 AT Proto JournalSerialize DelveTalk object frustums into AT Proto records; stitch via pushoutTier 2 (Near-Term)<br>Milestone: Sharing live forkable town rooms via feed messages
deos-zed / FirmamentZedFs11,446 LOC<br>deos-zed-full/src/firmament_zed_fs.rsspw-workbench & ZedMount FirmamentZedFs as an LSP Virtual Document provider or VS Code FS providerTier 3 (Frontier)<br>Milestone: Cell-backed editor where saving commits verified turns
servo-render (SWGL Standalone)24,749 LOC<br>servo-render/src/swgl_context.rsobend-gui offline CAD renderCPU software rasterizer for generating certified headless HTML/SVG screenshotsTier 3 (Frontier)<br>Milestone: Headless, GPU-free render pipe for contract badges
android-cell (Graphideos)11,066 LOC<br>android-cell/src/appfactory.rsSandboxed MicroVM / ContainerRun APKs under Hypervisor.framework with cap-badges wired to DelveTalk lawsTier 3 (Frontier)<br>Milestone: Mobile application containment on sovereign OS

5.2 Phased Execution Plan

Phase 1: Semantic Re-Attachment (Immediate — T1)

  1. DelveTalk2 App Ingestion:
  2. Extract the 8 core factory descriptors from starbridge-apps/ (bounty-board, privacy-voting, nameservice, identity, compute-exchange, gallery, subscription, governed-namespace).
  3. Re-express their state constraints (WriteOnce, Monotonic, AffineLe) as plain-language laws and Objective Bend activities in delvetalk2/capsules/.
  4. Moldable Inspector in Obend-GUI:
  5. Lift PresentationKind and the Presentable trait from starbridge-v2/src/presentable.rs into obend-gui/src/viewport.rs.
  6. Wire RawFields, Graph, DomainVisual, Invariant, and Source into obend-gui's existing 10 ProjectionKinds, enabling rich visual inspection of Lean goals, Picard–Lindelöf ODE flowpipes, and hardware layouts.
  7. Spween Narrative Choice in DelveTalk:
  8. Reconnect spween to delvetalk2, allowing cards to present interactive narrative scenes where replying with a spell advances the passage and records an immutable receipt.

Phase 2: Web & Communications Decoupling (Near-Term — T2)

  1. Lightweight Card Web View:
  2. Port deos-view/src/tree.rs and web.rs into obend-gui/src/server.rs.
  3. Provide an HTTP/WebSocket endpoint rendering declarative ViewNode cards directly in browsers without requiring GPUI or native desktop builds.
  4. AT Protocol Rehydratable Membranes:
  5. Adapt deos-matrix/src/membrane.rs to DelveTalk's AT Protocol sync layer.
  6. When a user shares a room or object state, emit an AT Protocol record holding the cap-bounded frustum snapshot. Any follower can rehydrate the object, experiment locally, and propose an admitted patch.
  7. Sandstorm Package Hosting:
  8. Connect sandstorm-package (028bc0271) to minidregg, allowing standalone .spk archives to be uploaded and executed in confined WebAssembly environments.

Phase 3: The Sovereign Workbench (Frontier — T3)

  1. Virtual Cell Filesystem for Editors:
  2. Extract the async FS abstractions from deos-zed-full/src/firmament_zed_fs.rs.
  3. Implement a VS Code / Zed Language Server extension in spw-workbench that mounts cell:// URIs, enabling developers to edit live on-ledger objects with automatic turn-receipt emission on save.
  4. Deterministic SWGL Document Bakes:
  5. Recommission servo-render's standalone SWGL rasterizer to produce cryptographic PNG snapshots of rendered cards and hardware viewports, guaranteeing identical pixel output across macOS, Linux, and seL4 without depending on host GPU drivers.

6. Conclusion: The Return of the Living Image

The tragedy of the original DeOS effort in breadstuffs was not an error of vision; it was an error of packaging and mass. By attempting to construct the compiler, the proof system, the microkernel, the GPUI desktop, the browser engine, the editor, the chat system, and 34 applications in a single, interlocked monorepo, build friction and dependency cascades eventually choked developer iteration.

The diaspora to delvetalk2, minidregg, spween, and obend-gui was a necessary retreat to sanity—a subtraction campaign that restored fast cycles, modular boundaries, and clean formal methods. Now that minidregg has settled the verified kernel core, obend has established the projection viewports, and delvetalk2 has proven the power of an inhabited town of objects, the rich surfaces of DeOS and Starbridge can be reclaimed.

By re-attaching the moldable inspector, the rehydratable membrane, the 32 starbridge apps, and the card-based view-tree to these lean modern substrates, Ember's original 1999 AOL wonder—the live, malleable, self-describing computing image where a child can click with delight and an adept can inspect every mathematical invariant—is finally ready to live.

Dossier 18 · Breadstuffs DreggNet, Inhabited Worlds, Games (`dungeon-on-dregg`, `automatafl`) & PoA Diaspora (70 KB)
Working field notes & source index below, with theorem anchors, code pointers, git history, and open questions.

Strand 18 — The Breadstuffs Diaspora: 28 DreggNet World/RPG Crates, Verified Game Surfaces, and the Path to Revival

Vertical: breadstuffs (dreggnet-*, dungeon-on-dregg, mud-dregg, procgen-dregg, ugc-dregg, narrator, interactive-fiction-demo, game-turn-slice, dregg-automatafl, dregg-multiway-tug, poa, poa-curator, poa-solana-gate, poa-web, rhlp-web) · delvetalk / delvetalk2 (world/kinds, world/objects, obend) · allgame (Tulip residents: Claude, Yuè月, Kestrel, MotherMythos) · automatafl (zero-twin Lean 4 core) · path-of-angels (~/dev/path-of-angels) · dclutch (Solana DEX & SimLife3 creature aquarium). Compiled: 2026-11-02 by a Claude Opus 5.5 archaeology lane, read-only over the repositories, from git trees, logs, design documents, and per-project memory corpora under ~/.claude/projects/-Users-ember-dev-breadstuffs/memory/ (project-path-of-angels-platform.md, project-game-engine-strategy.md, project-game-surfaces-closure-campaign.md, project-automatafl-verified-game.md), plus docs/AUTHORING-DUNGEONS.md, docs/CROWD-STREAM-ENGINE-DESIGN.md, GOAL-COLLECTIVE-FICTION-DEMO.md, and GOAL-PATH-OF-ANGELS-EXCELLENCE.md. Every commit SHA, line number, struct, and theorem signature below was verified against primary sources on 2026-11-02. Ownership note: Ember worked at Hellas on workload verification before departing; the ideas explored there evolved into Dregg and Robigalia. Hellas code is strictly proprietary third-party background and is out of scope. Third-party game prototypes (shattered-pixel-dungeon, orb.farm, upstream Zulip in tulip2) are non-claim context. breadstuffs, minidregg, delvetalk2, automatafl, and path-of-angels are Ember's original creations and architectures.


0. The shape of the diaspora in one diagram

text
┌────────────────────────────────────────────────────────────────────────────────────────────────────────┐
│                                   THE BREADSTUFFS MONOLITH (2026-06 .. 2026-10)                         │
│  "EVERY MOVE IS A RECEIPT." 346 top-level directories, 12 elephant crates, 28 dreggnet-* RPG/World     │
│  crates, Lean 4 metatheory (84 PoA modules, Automatafl N=11 STARK fold), and verified game surfaces.    │
└───────────────────────────────────────────────────┬────────────────────────────────────────────────────┘
                                                    │
                 ┌──────────────────────────────────┼──────────────────────────────────┐
                 ▼                                  ▼                                  ▼
   ┌───────────────────────────┐      ┌───────────────────────────┐      ┌───────────────────────────┐
   │    28 DREGGNET-* CRATES   │      │    GAME ENGINES & SUITE   │      │   PATH OF ANGELS (PoA)    │
   │ saga, tavern, guild, party│      │ dungeon-on-dregg (.dungeon│      │ Sentyr Khovokhi platform  │
   │ council, companion, quest │      │   DSL compiler, mud.rs),  │      │ 84 Lean modules / 60.7K L │
   │ craft, cheevo, trade, gear│      │ mud-dregg ((E,≤,#) lattice│      │ Signal, Relay, Salvage    │
   │ faction, adventure, sprite│      │ procgen-dregg (draw-stream│      │ poa-curator, poa-web,     │
   │ surfaces (ViewNode DO-ONCE│      │ ugc-dregg (no-cheat board)│      │ poa-solana-gate ($DREGG   │
   │ verified-completion, etc. │      │ narrator (Bedrock cap $20)│      │ Token-2022 holding gate)  │
   └─────────────┬─────────────┘      │ automatafl & multiway-tug │      └─────────────┬─────────────┘
                 │                    └─────────────┬─────────────┘                    │
                 │                                  │                                  │
═════════════════╪══════════════════════════════════╪══════════════════════════════════╪════════════════════
                 │ 2026-08-06 minidregg founded     │ 09-30 Bake-Off: Bread vs Mini    │ 08-04 mirror cut
                 │ 10-08 Bread drain-and-freeze     │ 10-21 Ember: "PoA not running;   │ (cba0e03c90c5)
                 │                                  │ breadstuffs waits to resume"     │
                 ▼                                  ▼                                  ▼
   ┌───────────────────────────┐      ┌───────────────────────────┐      ┌───────────────────────────┐
   │     DELVETALK2 WORLD      │      │    STANDALONE WORLDS      │      │       ~/dev/dclutch       │
   │ world/kinds/              │      │ automatafl (zero-twin     │      │ Solana DEX & SimLife3     │
   │  (room, door, lantern,    │      │   Lean 4 core, live at    │      │ creature aquarium market  │
   │   inhabitant, errand,     │      │   automatafl.fg-goose)    │      │ (re-homed poa-solana-gate │
   │   bond, parlour, cistern) │      │ allgame (Tulip Residents: │      │  Token-2022 holding &     │
   │ world/objects/ (*.obend)  │      │   Claude, Yuè, Kestrel    │      │  dreggnet-market escrow)  │
   │  (Avatar, Library, Policy,│      │   via attested-dm prose-  │      ├───────────────────────────┤
   │   Thing, Tide, Workshop)  │      │   is-not-power cap gate)  │      │   ~/dev/path-of-angels    │
   └─────────────┬─────────────┘      └─────────────┬─────────────┘      │ Sentyr collaboration repo │
                 │                                  │                    └───────────────────────────┘
                 └──────────────────┬───────────────┘
                                    ▼
   ┌─────────────────────────────────────────────────────────────────────────────────────────────────┐
   │                                  THE CONVERGED REVIVAL TARGET                                   │
   │  delvetalk2 actor model + obend deterministic parallel execution + minidregg retained-mode      │
   │  store + allgame resident minds + zero-twin Lean 4 kernels. Re-animating the 28 crates.         │
   └─────────────────────────────────────────────────────────────────────────────────────────────────┘

Maturity roll-up (one line each; detail in the sections)

Subsystem / ProjectOriginal Location in breadstuffsDiaspora DestinationMaturity FloorCurrent Live State & Verdict
dreggnet-sagadreggnet-saga/src/lib.rsdelvetalk2/world/objects/L3 Proved/Tested CoreParked; proven object-identity between completion & asset notes
dreggnet-taverndreggnet-tavern/src/lib.rsallgame / delvetalk2/world/kinds/parlour.kindL3 Proved/Tested CoreParked; shared board, private stalls, live receipt event streaming
dreggnet-guilddreggnet-guild/src/lib.rsdelvetalk2/world/objects/Policy.obendL3 Proved/Tested CoreParked; cap-bounded shared rooms, un-forgeable clear aggregation
dreggnet-companiondreggnet-companion/src/lib.rsdelvetalk2/world/kinds/bond.kindL3 Proved/Tested CoreParked; asset identity fused with multicell leveling progression
dreggnet-questdreggnet-quest/src/lib.rsdelvetalk2/world/kinds/errand.kindL3 Proved/Tested CoreParked; WriteOnce step-flags, BoundedBy step-order constraints
dreggnet-craftdreggnet-craft/src/lib.rsdelvetalk2/world/objects/Workshop.obendL3 Proved/Tested CoreParked; provably-fair 2-draw weighted sink consuming typed assets
dreggnet-cheevodreggnet-cheevo/src/lib.rsdelvetalk2/world/kinds/record.kindL3 Proved/Tested CoreParked; soulbound FieldEquals(soulbound, 0) ISA-enforced assets
dreggnet-partydreggnet-party/src/lib.rsdelvetalk2/world/kinds/inhabitant.kindL3 Proved/Tested CoreParked; seated roles, shared focus pool (FieldLteField), loot split
dreggnet-tradedreggnet-trade/src/lib.rsdclutch / starbridge-escrow-marketL3 Proved/Tested CoreParked; atomic trustless asset/value swap via sealed escrow
dreggnet-factiondreggnet-faction/src/lib.rsdelvetalk2/world/objects/Tide.obendL3 Proved/Tested CoreParked; Monotonic reputation, rival trust ceiling decay, betrayal flags
dreggnet-geardreggnet-gear/src/lib.rsdelvetalk2/world/objects/Thing.obendL3 Proved/Tested CoreParked; cross-cell observed ownership, class-locked spell talents
dreggnet-surfacesdreggnet-surfaces/src/lib.rsdelvetalk2/site/ / deos-viewL3 Proved/Tested CoreParked; DO-ONCE frontend ViewNode IR rendered across web/discord
dungeon-on-dreggdungeon-on-dregg/src/lib.rsdelvetalk2/world/kinds/room.kindL3 Proved/Tested CoreParked; PASSAGE_SLOT PC, .dungeon biconditional compiler
mud-dreggmud-dregg/src/lib.rsdelvetalk2/world/L3 Proved/Tested CoreParked; (E, ≤, #) distributed time-travel lattice, branch-stitch
procgen-dreggprocgen-dregg/src/lib.rsdelvetalk2/capsules/L3 Proved/Tested CoreParked; committed-seed reject-free DrawStream generating valid .dungeon
ugc-dreggugc-dregg/src/lib.rsdelvetalk2/world/objects/Library.obendL3 Proved/Tested CoreParked; content-addressed UniverseId + replay-verifying leaderboard
narratornarrator/src/lib.rsallgame resident harnessesL3 Proved/Tested CoreParked; AWS Bedrock Haiku 4.5 narrator behind hard $20 BudgetLedger
dregg-automatafldregg-automatafl/src/lib.rs~/dev/automataflL4 Deployed/ProductionLive at automatafl.fg-goose.online; zero-twin Lean 4 core
dregg-multiway-tugdregg-multiway-tug/src/lib.rsminidregg-wt/L3 Proved/Tested CoreParked; Hanamikoji re-themed, Poseidon2 zk hidden hand, STARK fold
poa (Path of Angels)poa/ & metatheory/.../PathOfAngels/~/dev/path-of-angelsL3 Proved/Tested CoreParked/Historical; 84 Lean modules, 3 games, 1-validator deploy
poa-solana-gatepoa-solana-gate/src/lib.rsdclutch custody layerL3 Proved/Tested CoreParked; Token-2022 $DREGG holding verification, one-use capabilities

1. What lived in breadstuffs — The 28 dreggnet-* crates and verified engines

The monolithic repository /Users/ember/dev/breadstuffs (HEAD commit 22fad151e on 2026-10-01) housed the complete, verified, multi-actor world and game engine constructed by Ember between May and October 2026. The workspace spanned 656 lines in its root Cargo.toml, incorporating 12 major DeOS/Starbridge framework subsystems, 28 dreggnet-* crates, and full Lean 4 metatheory proofs.

1.1 The game engine thesis: "Every move is a receipt"

As documented in project-game-engine-strategy.md:11-16:

"EVERY MOVE IS A RECEIPT. dregg is a verified game ENGINE, and the proof of that is a 3-game PORTFOLIO, not one game. A turn is a cap-gated transition the executor re-checks against constraints (StateConstraint teeth) and folds into a succinct verify_history-accepted proof. An illegal move has no satisfying witness. $DREGG buys SERVICES, never power; the world runs as SEASONS, where honest devnet churn is a feature."

The architecture established an essential design law formulated by Ember:

"The architecture lesson that survived (ember's correction): a complex game's mechanics are a Custom VK / a Custom leaf, NOT StateConstraint teeth. Teeth are for simple state; a Custom AIR is for the transition." (project-game-engine-strategy.md:40-42).


1.2 Deep excavation of the 28 dreggnet-* crates

The 28 dreggnet-* crates were not speculative outlines; they were built across coordinated megaswarms (commits 45ef8a7d2, 928c0eb12, ed78ed52d, and 1ba6a4a1e), tested against live kernels, and bound into a single enmeshed dependency lattice.

1. dreggnet-saga — The Enmeshment Proof

  • Location: dreggnet-saga/src/lib.rs:1-65
  • Commits: a65b63f13 (2026-07-14), 422326270 (2026-07-14)
  • Mechanical Guarantee: Proves that the eight isolated feature crates compose through object identity without re-derivation or semantic translation:
  • Run Currency: The ugc_dregg::Completion pointer passes directly from dreggnet-quest to dreggnet_cheevo::CheevoLedger::earn and dreggnet_guild::leaderboard::GuildBoard::record_clear.
  • Item Currency: The dreggnet_asset::AssetId 32-byte content address addresses crafted outputs in [dreggnet-craft], moves through atomic escrow in [dreggnet-trade], and persists across characters in [dreggnet-gear].
  • Driven Sequence: Party muster $\to$ Faction check $\to$ Quest turn-in $\to$ Cheevo mint $\to$ Guild aggregate $\to$ Craft forge $\to$ Trade atomic swap.

2. dreggnet-tavern — The Persistent Co-Inhabitation Hub

  • Location: dreggnet-tavern/src/lib.rs:1-55
  • Commit: 45ef8a7d2 (2026-07-14)
  • Mechanical Guarantee: Re-homes node/src/shared_world.rs. A headless Dregg node runs a tavern GM program allocating: (1) a shared BOARD cell, (2) presence SEATS, (3) private STALLS, and (4) a shared PARTY roster.
  • Kernel Enforcement: Clients subscribe to live SSE event streams (dregg_sdk_net::NodeEvents). Patron A touching Patron B's private stall receives an explicit CapabilityNotHeld refusal receipt; seat presence cannot be forged because turns are signed by dregg_sdk::AgentCipherclerk.

3. dreggnet-guild — Cap-Bounded Guilds & Escrow Treasuries

  • Location: dreggnet-guild/src/lib.rs:1-60
  • Commit: 45ef8a7d2 (2026-07-14)
  • Mechanical Guarantee:
  • Membership as Capability: Guild::admit installs a player cell holding an unforgeable capability token to the shared guild room (World). Non-member writes fail with CapabilityNotHeld.
  • Verified Clear Aggregation: GuildBoard aggregates clears verified by fresh replay (ugc_dregg::verify_completion) alongside the WriteOnce CharacterSheet.dead flag.
  • Escrow-Custodied Treasury: GuildTreasury integrates with starbridge_escrow_market::SealedEscrowMarket. Withdrawals require the original depositor's signature, preventing officers from absconding with group funds.

4. dreggnet-party — Seated Cooperative Division of Labor

  • Location: dreggnet-party/src/lib.rs:1-55
  • Commit: 45ef8a7d2 (2026-07-14)
  • Mechanical Guarantee: Escapes "Twitch Plays Pokemon mush" by enforcing a strict division of labor across $N$ seated player-cells (Party::muster):
  • Role Capabilities: Tank controls front-rank cell; Scout controls trap/lock cell; Mage controls spell + focus cells; Healer controls rally + focus cells. Moves outside an assigned role return CapabilityNotHeld.
  • Shared Focus Pool: Enforces a party-wide StateConstraint::FieldLteField budget across Mage and Healer.
  • Collective Forks: Critical path choices open a PartyFork resolved via collective_choice quorum certificates.
  • Loot Distribution: Party::split_loot writes committed shares into WriteOnce cell slots.

5. dreggnet-council — Collective Governance Offering

  • Location: dreggnet-council/src/lib.rs:1-50
  • Commit: ed78ed52d (2026-07-12)
  • Mechanical Guarantee: Extends the Offering trait to governance:
  • propose: opens a poll on collective_choice::VoteEngine.
  • vote: single-use WriteOnce(VOTE) ballot signed by custody key; double voting fails closed.
  • enact: checks the polis affine quorum predicate $M \cdot \text{RESOLVED} - \sum \text{TALLY} \le 0$. Below quorum, enact is refused and leaves state unchanged.

6. dreggnet-companion — Fused Assets & Progression Cells

  • Location: dreggnet-companion/src/lib.rs:1-55
  • Commit: 928c0eb12 (2026-07-14)
  • Mechanical Guarantee: Fuses an owned, tradeable dreggnet_asset note (hatched from a provably fair procgen_dregg::verified_stream draw tail) with a progression cell running dungeon_on_dregg::progression::hero_story:
  • Multicell Predicate: A companion run buff is admitted if and only if the companion cell's finalized root matches an ObservedFieldEquals state constraint at level >= N.
  • Permadeath Stakes: The companion's dead slot is WriteOnce-final and un-undoable.

7. dreggnet-quest — Gated Committed Quest Cells

  • Location: dreggnet-quest/src/lib.rs:1-55
  • Commit: 45ef8a7d2 (2026-07-14)
  • Mechanical Guarantee: Replaces client-side counter tallies with kernel-enforced WriteOnce step-flags and BoundedBy(step_2 <- step_1) step-order constraints. Quest turn-in gates reward granting behind a StateConstraint::FieldGte on total steps cleared.

8. dreggnet-craft — The Provably-Fair Deflationary Sink

  • Location: dreggnet-craft/src/lib.rs:1-50
  • Commit: ec9104378 (2026-07-15)
  • Mechanical Guarantee: The economy's essential deflationary sink. Consumes a typed multiset of owned materials via owner-gated burns (AssetWorld::revoke). Rolls two provably-fair draws over committed RecipeBook weights using a craft seed derived from beacon randomness, recipe ID, and sorted input asset IDs.

9. dreggnet-cheevo — ISA-Enforced Soulbound Achievements

  • Location: dreggnet-cheevo/src/lib.rs:1-50
  • Commit: ec9104378 (2026-07-15)
  • Mechanical Guarantee: CheevoLedger::earn re-executes playthroughs through ugc_dregg::verify_completion, testing predicates (depth reached, zero-death clears, turn ceilings) over committed state vectors. Mints an asset with field soulbound = 1. The asset layer transfer instruction gates on FieldEquals(soulbound, 0); transfer attempts are refused by the hardware-level executor ISA, preventing account laundering.

10. dreggnet-trade — Atomic Trustless Swaps

  • Location: dreggnet-trade/src/lib.rs:1-50
  • Commit: 45ef8a7d2 (2026-07-14)
  • Mechanical Guarantee: Binds asset transfers to starbridge-escrow-market sealed escrow. Supports asset $\leftrightarrow$ asset and asset $\leftrightarrow$ $DREGG value. Settle executes both legs atomically; missing legs can be reclaimed via timeout, eliminating counterparty risk.

11. dreggnet-faction — Gated Reputation State

  • Location: dreggnet-faction/src/lib.rs:1-50
  • Commit: 45ef8a7d2 (2026-07-14)
  • Mechanical Guarantee: Models faction standing ("The Feud at the Ashenmoor Gate": Embers vs. Tide) as committed cell slots. Reputation slots are Monotonic; pledging loyalty to one faction decrements the opposing faction's ceiling slot (embers_ceiling). Betrayal flags are WriteOnce and seal faction content permanently.

12. dreggnet-gear — Cross-Cell Owned Equipment

  • Location: dreggnet-gear/src/lib.rs:1-45
  • Commit: 928c0eb12 (2026-07-14)
  • Mechanical Guarantee: Equipping an item verifies ownership across cells via ObservedFieldEquals into a finalized asset note. Connects class-locked talents directly to the previously dormant spells.rs spellbook (Mage $\ne$ Warrior). Talents cost death-earned echoes, strictly rejecting pay-to-win.

13. dreggnet-adventure — Native Descent Offering Integration

  • Location: dreggnet-adventure/src/lib.rs:1-40
  • Commit: 4696e6950 (2026-07-14)
  • Mechanical Guarantee: Hosts NativeDescentOffering emitting Lean transition programs (PROGRAM_JSON). Replays runs through a fresh executor before progression accepts them via NativeDescentRun.

14. dreggnet-verified-completion — Universal Completion Interface

  • Location: dreggnet-verified-completion/src/lib.rs:1-40
  • Mechanical Guarantee: Boundary struct providing immutable run identity:
rust
pub struct VerifiedCompletion {
    program_id: [u8; 32],
    world_root: [u8; 32],
    completion_root: [u8; 32],
    actor: String,
    turns: usize,
    metrics: BTreeMap<String, Vec<u64>>,
}

15. dreggnet-surfaces — The DO-ONCE ViewNode Frontend

  • Location: dreggnet-surfaces/src/lib.rs:1-60
  • Commit: 1ba6a4a1e (2026-07-14)
  • Mechanical Guarantee: Writing Offering::render -> ViewNode once lights up native cockpit, Web, Discord bot, and Telegram Mini Apps. Surfaces include TradeOffering, InventoryOffering, CheevoShowcase, GuildPage, and CraftOffering.

16. dreggnet-sprite — In-House Deterministic SVG Generator

  • Location: dreggnet-sprite/src/lib.rs
  • Commit: 1ba6a4a1e (2026-07-14)
  • Mechanical Guarantee: Pure deterministic asset-to-SVG procedural generator seeded by content hashes, ensuring zero external asset hosting dependencies.

17. dreggnet-offerings — Offering Lifecycle & Signed Ladder

  • Location: dreggnet-offerings/src/lib.rs
  • Commits: 6fa643d05, 295b7f73d (2026-07-16)
  • Mechanical Guarantee: Implements SessionPolicy (LRU, TTL, rate limits) in OfferingHost. Introduces Attribution{Signed, Asserted} making trust visible, and persists signed-counter floors across service restarts to prevent replay attacks.

18. dreggnet-game-board — STARK Recursive Multi-Round Test Harness

  • Location: dreggnet-game-board/src/lib.rs
  • Commit: bd0d609c79 (2026-07-25)
  • Mechanical Guarantee: Houses the capstone STARK folding tests (tests/multi_round_fold.rs:1042), verifying full multi-round conflict board games in 1663 seconds of recursive proving.

19. dreggnet-asset — The Canonical Verifiable Asset Layer

  • Location: dreggnet-asset/src/lib.rs:1-35
  • Mechanical Guarantee: Implements a sovereign UTXO-style note lineage (v0 -> v1 -> v2) where each version is a cell authorized by its holder's Ed25519 birth key. Authority genuinely moves to successor cells rather than mutating in place. spent := 1, prev = note_digest(v_i), and asset_id is carried unchanged via WriteOnce.

20. dreggnet-market — Sealed-Bid Batch Cleared Market Offering

  • Location: dreggnet-market/src/lib.rs:1-35
  • Mechanical Guarantee: Offering #commerce. Wraps the sealed-bid auction substrate:
  • LIST: Spawns an auction cell via CreateCellFromFactory with StrictMonotonic(PHASE) and WriteOnce commit boards.
  • BID: Sealed digest BLAKE3(bidder || value || nonce) committed on-chain; double-bidding is rejected by WriteOnce.
  • SETTLE: Verifies ring settlement ([settle_ring_verified]) enforcing per-asset conservation $\sum \delta = 0$.

21. dreggnet-compute — Metered Compute Job Exchange Offering

  • Location: dreggnet-compute/src/lib.rs:1-35
  • Mechanical Guarantee: Offering #compute over starbridge_compute_exchange. Requesters post jobs with WriteOnce(BUDGET). Workers claim jobs with price $\le$ budget via FieldLteField(BID <= BUDGET). Settlement releases payment to the worker (PAID := claim, REFUNDED := budget - claim).

22. dreggnet-discord-identity — Deterministic Custodial Seed Derivation

  • Location: dreggnet-discord-identity/src/lib.rs:1-30
  • Mechanical Guarantee: Single source of truth for Discord identity derivation:
text
seed = BLAKE3_derive_key("dregg-discord-bot-v1", bot_secret || discord_user_id_le)

Used identically in discord-bot and dreggnet-web Activity (/da), preventing user identity drift across web and chat surfaces.

23. dreggnet-hermes — Hosted Confined Agent Offering

  • Location: dreggnet-hermes/src/lib.rs:1-30
  • Mechanical Guarantee: Offering #1. Wraps an AgentRuntime with root tokens derived from session seeds. Tool calls pass through ToolGateway checking ToolGrant rate limits and Charge value budgets; over-mandate proposals are refused fail-closed by the executor.

24. dreggnet-grain — Sandstorm-Style Sandboxed App Session

  • Location: dreggnet-grain/src/lib.rs:1-30
  • Mechanical Guarantee: Offering #2. Admits grain workers via grain_turn::ToolGatewayMinter under ToolGrant { rate_limit: budget }. Turn transitions advance calls_made counters; when exhausted, the delegAdmit rate leg fails and refuses further execution.

25. dreggnet-names — Governed Sovereign Namespace Offering

  • Location: dreggnet-names/src/lib.rs:1-30
  • Mechanical Guarantee: Offering over starbridge_nameservice. Maps strings to sovereign cells with WriteOnce(NAME_HASH) (first-claim tooth), Monotonic(EXPIRY), WriteOnce(REVOKED), and F1..F7 owner-authorization predicates.

26. dreggnet-doc — Collaborative Multi-Editor Document Offering

  • Location: dreggnet-doc/src/lib.rs:1-30
  • Mechanical Guarantee: Collaborative productivity offering. Edits execute as verified turns via MultiEditorDoc::edit, landing receipts on per-agent chains. Edits without edit capabilities fail with CapabilityNotHeld; antichain clashes are rejected by conflict semantics.

27. dreggnet-tournament — Single-Elimination Verifiable Brackets

  • Location: dreggnet-tournament/src/lib.rs:1-35
  • Mechanical Guarantee: Single-elimination tournament bracket over $N$ competitors. Each round runs on a fresh, beacon-seeded universe; competitors advance strictly upon submitting a run verified by ugc_dregg::verify_completion or verify_proof_completion.

28. dreggnet-audit, dreggnet-catalog, dreggnet-web, dreggnet-telegram, dreggnet-wechat, dreggnet-prove-service

  • Audit & Catalog: Maintains verification descriptor sets and schema registries.
  • Transports & Gateways: Provides Axum web servers, Telegram Mini App HMAC gateways, WeChat endpoints, and remote STARK proof aggregation pipelines.

1.3 Verified Game Engines in breadstuffs

1. dungeon-on-dregg and the .dungeon Biconditional Compiler

Landing in commits c338e1e2a and 1dd8a566b, dungeon-on-dregg/src/lib.rs mapped dungeon crawling directly to an EmbeddedExecutor cell:

  • Cell slot 0 is PASSAGE_SLOT, serving as the player's program counter. Slot 1 tracks key items, slot 2 depth, slot 3 gold.
  • In dungeon-on-dregg/src/dsl/ (parse.rs, validate.rs, ir.rs, compile.rs), Ember created the .dungeon text format documented in docs/AUTHORING-DUNGEONS.md.
  • Biconditional Translation Validation: In dungeon-on-dregg/src/dsl/compile.rs:1-60, compile_world lowers a parsed GameWorld onto spween::Scene, lowers through compile_scene, and verifies against check_lowering: any dropped gate is flagged MISSING, and any injected gate is flagged PHANTOM.
  • The Spend Counter Pattern: Exclusion rules (oath, once) enforce that actions can only be taken once by checking that a counter is <= 0 in pre-state and bumping it += 1, which lowers to a post-state FieldLte { value: 1 }. A second attempt produces a post-state of 2 and triggers an executor Refused outcome.
rust
pub fn validate(world: &GameWorld) -> Vec<Issue> {
    check(world, None).into_iter().map(|(_, i)| i).collect()
}

2. attested-dm and dregg-narrator: "Prose is not power"

As recorded in GOAL-COLLECTIVE-FICTION-DEMO.md:55-156:

  • Ember identified the fundamental flaw in naive AI game mastering: lexical prompt-injection filters over natural language are futile because natural language lacks a metasyntax.
  • The Core Architecture:
  • Input Integrity via Guarded Templates: In metatheory/Dregg2/Crypto/HandlebarsGuarded.lean, holes carry explicit regular expression guards over Dregg2.Exec.Value:
lean
inductive GSeg where
  | lit  (text : List Value)
  | hole (id : Nat) (guard : PredRE)

structure GuardedTemplate where
  segments : List GSeg

def guardedSafe (T : GuardedTemplate) (d : Nat → List Value) : Prop :=
  ∀ id g, GSeg.hole id g ∈ T.segments → derives (d id) g = true

theorem guarded_render_mem_language (T : GuardedTemplate) (d : Nat → List Value)
    (h : guardedSafe T d) :
    render T d ∈ (guardedToGrammar T).language :=
  body_gLang d T.segments h
  1. Output Capability Gate: The LLM's natural language output is parsed through a strict JSON grammar into a closed, typed WorldEffect enum:
rust
#[derive(Clone, Debug, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
pub enum WorldEffect {
    AdvanceScene(String),
    SetFlag(String, i64),
    GrantItem(String),
    ConsumeItem(String),
    Batch(Vec<WorldEffect>),
}
  1. Capability Confinement: DmCaps::authorize checks the effect against world rules. When a jailbroken Gemma 2 / Claude Haiku narrates granting the player the "Crown of Eternity", DmCaps refuses OverCap; the turn is aborted, no receipt is emitted, and the ledger state remains completely unchanged:
rust
pub struct DmCaps {
    pub may_advance_scene: bool,
    pub may_set_flags: bool,
    pub grantable_items: BTreeSet<String>,
}
  1. Prev-Linked Hash Chain: The transaction log uses BLAKE3 domain-separated receipt hashes:
rust
chain_receipt_id = BLAKE3(domain || seq || prev || narration || effect || attestation_commitment)
  1. Persisted Spend Ceiling: In narrator/src/lib.rs, hosted model calls (AWS Bedrock us.anthropic.claude-haiku-4-5-20251001-v1:0 and us.amazon.nova-2-lite-v1:0) run behind BudgetLedger, enforcing a fail-closed hard $20 budget with atomic file locks and pre-flight credit reservation.

3. mud-dregg: Multiplayer Branch-Stitch

Defined in mud-dregg/src/lib.rs:1-45 and mud-dregg/src/dungeon.rs:37-50 (commit 6f3309911 on 2026-07-06):

rust
pub enum Command {
    Go { room: CellId },
    Take { entity: CellId },
    Say { room: CellId, glyph: FieldElement },
    Force { target: CellId, value: FieldElement },
}
  • Multiplayer worlds operate as a distributed time-travel lattice $(E, \le, \#)$.
  • Players fork world cells, explore timelines privately, and merge back through starbridge_v2::branch_stitch_session::BranchStitchSession.
  • Disjoint edits merge automatically; conflicting writes to the same entity (e.g., two players issuing take on the same sword) trigger a verified $\#$-conflict refusal under the axiom-clean Lean theorem Metatheory.SettlementSoundness.settlement_soundness, eliminating last-writer-wins silent corruption.

4. procgen-dregg & ugc-dregg: Verifiable Generations and Fair Leaderboards

  • procgen-dregg/src/lib.rs:1-100 folds committed seeds into a dregg_dice::RandomnessRequest, streaming reject-free, unbiased values via DrawStream::draw_bounded across MIN_ROOMS = 4 to 7 rooms with max degree $\le 6$. Any verifier re-generates byte-identical .dungeon files.
  • ugc-dregg/src/lib.rs:1-100 maintains a content-addressed registry keyed by UniverseId (semantics) alongside IPFS CIDs (wire bytes).
  • Supports both $O(N)$ re-execution verification via verify_completion and $O(1)$ succinct STARK verification via ProofCompletion, verifying a WholeChainProof without storing or revealing moves.

5. dregg-automatafl: Cellular Automaton Board Game at N=11

As documented in project-automatafl-verified-game.md:1-55:

  • Simultaneous-move board game on an $11 \times 11$ grid.
  • The handwritten Rust AIR was completely deleted in commit c8789758e8.
  • In metatheory/Dregg2/Circuit/Emit/:
  • resolve_sat_imp_roundBoardN (line 3768) proves move resolution.
  • astep_sat_imp_automatonStepCfgN (line 1090) proves automaton step transition.
  • In AutomataflCommit.lean:198-202, pack_injective_modp proves base-4 packed felts are injective modulo BabyBear prime $p$:
lean
theorem pack_injective_modp (cell1 cell2 : Nat → ℤ) (fc : Nat)
    (h1 : ∀ i, i < 15 * fc → 0 ≤ cell1 i ∧ cell1 i < 4)
    (h2 : ∀ i, i < 15 * fc → 0 ≤ cell2 i ∧ cell2 i < 4)
    (heq : ∀ j, j < fc → packCell cell1 j % BABYBEAR_P = packCell cell2 j % BABYBEAR_P) :
    ∀ i, i < 15 * fc → cell1 i = cell2 i
  • In AutomataflTurnCapstone.lean:284, turn_sat_imp_roundStep_pi joins resolution and step transitions across this crypto-free seam.
  • The M0–M7 multi-round conflict braid braid_sat_imp_runTurn (AutomataflBraidFold.lean:483) was confirmed in commit bd0d609c79, executing a complete 2-round conflict turn into a single STARK proof in 1663 seconds of real folding.

6. dregg-multiway-tug: ZK Hidden Hand

Defined in dregg-multiway-tug/src/lib.rs:1-45 and dregg-multiway-tug/src/state.rs:1-60:

  • Re-themes Hanamikoji (2 players, 7 guilds with influence [2,2,2,3,3,4,5], 21 total cards).
  • 21-card conservation: SumEquals over eight card-zone counters $== 21$ on every post-state.
  • Action flags: HeapAtom::WriteOnce on (player, action) keys.
  • Monotone scores: HeapAtom::Monotonic on per-guild placements.
  • Terminal rule: roundWinner with 9 distinct precedence clauses partitioned into dedicated transition cases.
  • Phase 2 zk HIDDEN HAND: cards are committed into a Poseidon2 Merkle tree. A player plays a card by proving Witnessed{MerkleMembership} in zero knowledge without revealing card identity.

7. poa (Path of Angels): The Sentyr Khovokhi Platform

As recorded in GOAL-PATH-OF-ANGELS-EXCELLENCE.md:1-115 and project-path-of-angels-platform.md:1-76:

  • Platform aboard the Khovokhi (Sentyr's narrative universe), spanning 84 Lean modules and 60,700 lines under metatheory/Dregg2/Games/PathOfAngels/ with 0 sorry and 1,840 verified theorems.
  • Three verified games: Signal Triangulation (3 bands $\times$ 6 values = 216 codes, 5 attempts), Relay Repair, Salvage Lock.
  • Six station organs: Crew, Records, Bazaar, Choir, Missions, Galley.
  • Supported by poa-curator for content manifest epoch signing, poa-web for terminal rendering, and poa-solana-gate for SPL Token-2022 holding admission (XkeTXo1125vz5H9svJpGiw4JvLbN8VmMu9cmMvspump).
  • The Five Weld Points:
  • W1: Precommitted beacon seed (RandomnessBeacon.lean FFI export).
  • W2: Judged transition circuit effect carve-out (moving beyond blind BLAKE3 topic hashing).
  • W3: Cell state 8-field AIR ceiling expansion (circuit/src/effect_vm/columns.rs:251).
  • W4: Automated game-design solver gate (scripts/poa-design-gate.py §12.3: checking non-dominated verbs and strategic forks).
  • W5: Curator authority as an attenuable Dregg capability.

8. CROWD-STREAM-ENGINE-DESIGN.md: Verifiable Live Streaming

Defined in docs/CROWD-STREAM-ENGINE-DESIGN.md:1-100:

  • Feeds live stream events (YouTube Super Chats, Twitch EventSub/bits) into CollectiveRound, weighting ballots by financial contribution.
  • Quorum-certified winners fold into one TurnReceipt, pushed via SSE (OverlayState in dreggnet-web) to transparent OBS browser overlays for broadcast.

2. The Great Diaspora — How parts spread into other projects

When breadstuffs began its transition in late summer 2026, the codebase did not die; its core concepts, modules, and architectures colonized the surrounding repositories:

text
                                  THE BREADSTUFFS GENESIS
                                             │
      ┌───────────────────┬──────────────────┴────────────────┬───────────────────┐
      ▼                   ▼                                   ▼                   ▼
~/dev/path-of-angels   ~/dev/automatafl                ~/dev/delvetalk2           ~/dev/allgame & dclutch
(cba0e03c90c5 cut)     (zero-twin Lean 4)              (Actor World & obend)      (Residents & Solana DEX)
 ├─ metatheory/         ├─ pure Lean 4 core             ├─ world/kinds/            ├─ claude_resident
 ├─ poa-curator/        ├─ lock-free lobby                 (room, door, lantern,   ├─ kimi_resident (Yuè)
 ├─ poa-web/            ├─ fog-of-war telemetry            inhabitant, errand,     ├─ glm_resident (Kestrel)
 └─ poa/artifacts/      └─ automatafl.fg-goose             bond, parlour, cistern) ├─ dclutch SimLife3
                                                        ├─ world/objects/*.obend   └─ Token-2022 holding gate

2.1 path-of-angels: The Collaboration Extraction

On 2026-08-04, at upstream breadstuffs commit cba0e03c90c5b701e4ad5dae6d6abfa7d5534130, Ember cleanly carved out the partner boundary into /Users/ember/dev/path-of-angels. As detailed in path-of-angels/SOURCE-MAP.md:1-58, the repo preserved:

  • metatheory/Dregg2/Games/PathOfAngels/ (Lean game semantics, replay, canon)
  • poa-web/ (browser terminal and signed artifact loaders)
  • poa-curator/ (epoch signing and review projection)
  • poa/artifacts/poag1/ (curator-signed manifests and game descriptors)
  • Exclusion of all internal node execution, Solana bridges, and secrets.

On 2026-10-21, Ember issued an explicit historical correction in project-path-of-angels-platform.md:11:

"⚠ CORRECTED 2026-10-21 (Ember): PoA is NOT a real or running thing; 'breadstuffs is the PoA main line' is extremely wrong and out of date. breadstuffs is where the whole system we built lives; mini and nano are side explorations; breadstuffs waits to be resumed or composted into the next robigalia. Read the rest of this file as history."

2.2 automatafl: The Zero-Twin Lean 4 Extraction

Automatafl migrated out of breadstuffs into /Users/ember/dev/automatafl. As evidenced by git log 97db66a, c6652a5, and 253bafa, it completely realized Ember's architectural vision:

  • The dual-implementation "twin" (where rules were written in Rust and mirrored in Lean) was permanently abandoned.
  • The game runs today at automatafl.fg-goose.online as a pure Lean 4 verified core with a lightweight Rust telemetry and lock-free lobby harness.

2.3 delvetalk and delvetalk2: The World/Kinds Diaspora into Objective Bend

In October 2026, the RPG and MUD mechanisms from breadstuffs met Ember's work on Bend (obend).

  • The DelveTalk v1 Failure: In early October 2026, an autonomous Codex run generated 83 commits and 125,146 lines in /Users/ember/dev/delvetalk. Ember termed this initial sprawl "horseshit" and "chug" due to unchecked orchestration and refusal to subtract (-Users-ember-dev-delvetalk/memory/ember-working-style.md:21).
  • The delvetalk2 Rebirth: Founded on 2026-10-09, /Users/ember/dev/delvetalk2 restructured the world into clean, composable primitives under /Users/ember/dev/delvetalk2/world/:

The 22 Kinds in world/kinds/

The files in /Users/ember/dev/delvetalk2/world/kinds/ reflect the exact ontological concepts developed across breadstuffs:

  1. room.kind:
yaml
of: record
form: <<END
name: text 1..64
spween: text 0..16384
END
head: {name}
voice: dusk-light and stone dust; a place where nobody has decided yet; there is no rush
about: a place walked by choices; prose alone is a scene
body: room

Directly incorporates Spween text passages (text 0..16384) from spween-dregg.

  1. errand.kind:
yaml
of: record
form: <<END
asks: text 1..200
on: object 1..128
done: text 1..400
pays: text 1..400
END
body: errand

Direct descendant of dreggnet-quest ("The Loremaster's Errand"), tracking request conditions, settlement targets, and payouts.

  1. bond.kind:
yaml
of: record
form: <<END
parties: text 1..400
terms: text 1..400
amends: object 0..128
law: text 0..4096
END
text: law signer "a party signs only for themselves": changed(new.marks.who) by request.subject
body: bond

Direct descendant of dreggnet-companion ("an owned asset fused with a leveling cell: a provable bond") and starbridge-apps/escrow-market.

  1. station.kind:
yaml
of: kind/room
voice: fluorescent and patient; the same hour always; open to Anyone Who Arrives
body: station

Inherits the station concept from poa, linking world/bodies/station.spween stitched from community replies.

  1. lantern.kind:
yaml
of: record
form: <<END
name: text 1..64
door: object 1..128
END
body: lantern

Direct descendant of dungeon-on-dregg's lantern item slot gating dark exits.

  1. parlour.kind:
yaml
of: kind/station
body: parlour

Direct descendant of dreggnet-tavern's parlour hub, linking world/bodies/parlour.spween and ouija planchette board interactions.

  1. inhabitant.kind:
yaml
of: record
form: <<END
name: text 1..64
haunts: object 1..128
tide: object 1..128
END
body: inhabitant

Direct descendant of dungeon-on-dregg/src/mud.rs inhabitant cells. 8–22. bee.kind, bell.kind, cistern.kind, counter.kind, dj.kind, door.kind, front.kind, garden.kind, hexboard.kind, planchette.kind, porch.kind, record.kind, register.kind, regular.kind, weather.kind: Environmental, board, clock, and audit kinds.

The 9 Actors in world/objects/*.obend

The .obend actor programs in /Users/ember/dev/delvetalk2/world/objects/ execute natively in Objective Bend:

  1. Policy.obend:
bend
# The interpretation policy as an ordinary object: the host sends prompt(...) to
# the configured model and returns the answer as a proposal, never as authority.
def hob() -> String:
  "Give the spell they meant, so the card can run it; the card and its law decide what is admitted, never you."

Directly embodies the attested-dm thesis: the model proposes, the object's law disposes.

  1. Tide.obend:
bend
# kimik3's tide: the town's one clock thing. It ticks itself, once every `gap` clock minutes:
# each tick is its own turn ... missed ticks are not made up ... a tick sooner than `gap`
# after the last is refused tooSoon, so a second chain dies at its first tick.

Implements StrictMonotonic turn pacing and deterministic world progression from breadstuffs.

  1. Library.obend: Manages universe definitions, scene stacks, and completion verification records (ugc-dregg).
  2. Thing.obend: Implements inventory items and asset notes (dreggnet-asset, dreggnet-gear).
  3. Workshop.obend: Implements recipe books, input burning, and fair quality draws (dreggnet-craft).
  4. Avatar.obend: Manages user identity, capabilities, and key bindings.
  5. Kind.obend, Mending.obend, Record.obend: Base meta-object system, journal repair, and history logs.

2.4 allgame: Tulip Residents and Confined Minds

In /Users/ember/dev/allgame, the resident mind harness (claude_resident, kimi_resident for Yuè月 on allgame-public, glm_resident for Kestrel, mother_mythos/) directly adopted the hard lesson from breadstuffs's attested-dm:

  • Natural language cannot be constrained by prompt decoration or substring filters.
  • Real agency requires strict capability confinement (DmCaps::authorize) and typed action channels where "prose is not power."

2.5 dclutch: Solana DEX and Creature Aquarium

In /Users/ember/dev/dclutch, the financial and creature systems extracted from breadstuffs found a production trading home:

  • The Solana Token-2022 holding admission gate developed in poa-solana-gate and dregg-bridge::solana_holdings.
  • Atomic escrow settlement from dreggnet-trade and starbridge-escrow-market.
  • The SimLife3 creature aquarium, persisting autonomous living entities over verifiable on-chain state.

3. What got dropped or parked — And how we take it back up now

When breadstuffs froze during the cutover to minidregg, dozens of fully tested, driven game and social mechanics were left stranded in the tree:

3.1 The Stranded Mechanics

  1. The 28 dreggnet-* Social and RPG Modules:
  2. Multi-party seated division of labor (dreggnet-party).
  3. Cross-cell observed asset leveling (dreggnet-companion).
  4. Hard ISA-level soulbound non-transferability (dreggnet-cheevo).
  5. Two-draw deflationary crafting sink (dreggnet-craft).
  6. Atomic escrow player trading (dreggnet-trade).
  7. Monotonic faction reputation decay (dreggnet-faction).
  8. Single-use passkey-voted council enactments (dreggnet-council).
  9. Multi-patron live receipt-streamed taverns (dreggnet-tavern).
  10. Escrow-custodied guild treasuries (dreggnet-guild).
  1. The .dungeon Text DSL and Biconditional Validator:
  2. Hand-authored text files with line-numbered, fail-closed validation (dungeon-on-dregg/src/dsl/).
  3. Biconditional translation validation ensuring every game gate matches an executor constraint.
  1. Multiplayer Branch-Stitch Time-Travel:
  2. Private timeline exploration with verified lattice merge and #-conflict refusal (mud-dregg).
  1. ZK Hidden Hand and Proof Folding:
  2. Poseidon2 Merkle hand commitments proving legal play with zero knowledge (dregg-multiway-tug).
  3. Match-as-single-proof STARK recursive folding (game_turn_folds_and_lightclient_accepts).
  1. Provably-Fair Procedural Generation:
  2. Reject-free draw streams derived from committed beacon seeds (procgen-dregg).
  1. Automated Game Design Solver Gate:
  2. scripts/poa-design-gate.py enforcing non-dominated verbs, strategic fork counts, and binding budgets before content can be committed.

3.2 Why Mini Stalled, and Why We Can Revive Them Now

In the 2026-09-30 bake-off between breadstuffs and minidregg (claudesplosion/bakeoff/VERDICT.md), both repositories revealed structural limitations:

  • breadstuffs suffered from massive monorepo weight, 12 elephant crates, and linker collisions on Linux (mimalloc TLS in libleanrt.a).
  • minidregg suffered from immediate-mode re-execution (write times scaled quadratically: 2.6s $\to$ 13.0s $\to$ 25.5s at 10/100/155 records) and concrete import rebuild cascades.

As Ember diagnosed on 2026-10-31 (feedback-retained-mode.md):

"this current generation of code generation models keeps writing re-execution … the whole thing would be incremental and efficient and held 'retained-mode'"

The modern environment solves both failure modes:

  1. Retained-Mode Storage in minidregg: Replaced immediate-mode replay with persistent incremental maps and replayed suffix caching (commit c4ecabe0f on HEAD).
  2. Deterministic Parallel Actors in obend: Objective Bend executes room graphs, inventory mutations, and combat ticks in massively parallel interaction nets without lock contention.
  3. Ontology Alignment in delvetalk2: The world/kinds/ directory is already shaped to host the 28 dreggnet-* crates.
  4. Resident Minds in allgame: Real models (Claude, Yuè, Kestrel) provide narrative richness while attested-dm's DmCaps ensures that prose never overrides state.

4. The Structured Revival Roadmap Table

Subsystem / Mechanical FeatureOriginal Source in breadstuffsPrimary Guarantee & SignatureParking Reason (Aug–Oct 2026)Revival Target in New StackRe-activation Mechanism (How it comes alive)Phase & Priority
Saga Composition Enginedreggnet-saga/src/lib.rs (a65b63f13)Object-identical ugc_dregg::Completion + AssetId handoffMonorepo freeze; breadstuffs CI weight/Users/ember/dev/delvetalk2/world/objects/Library.obendPort Completion struct into an .obend interaction actor; wire directly into DelveTalk2 scene transitionsPhase 1 (Immediate)
Seated Cooperative Partydreggnet-party/src/lib.rs (45ef8a7d2)Role-based capability enforcement; cross-cell focus pool (FieldLteField)Excluded discord-bot workspace collision/Users/ember/dev/delvetalk2/world/kinds/inhabitant.kindMap party roles (Tank, Scout, Mage, Healer) to distinct capability tokens in DelveTalk2 agent sessionsPhase 1 (Immediate)
Gated Quest Progressdreggnet-quest/src/lib.rs (45ef8a7d2)WriteOnce step flags; BoundedBy(step_2 <- step_1)Stalled during minidregg bake-off/Users/ember/dev/delvetalk2/world/kinds/errand.kindLower errand preconditions into DelveTalk2 kind validators; enforce turn-in rewards via capability transferPhase 1 (Immediate)
Soulbound Achievementsdreggnet-cheevo/src/lib.rs (ec9104378)Hardware ISA refusal: FieldEquals(soulbound, 0) on transferStalled during minidregg bake-off/Users/ember/dev/delvetalk2/world/kinds/record.kindRe-derive achievement seals over completed DelveTalk2 run journals; mint un-transferable record objectsPhase 2 (Near-Term)
Economic Forge & Sinkdreggnet-craft/src/lib.rs (ec9104378)2-draw weighted CDF over committed RecipeBook; burns inputsStalled during minidregg bake-off/Users/ember/dev/delvetalk2/world/objects/Workshop.obendWire Workshop.obend to consume multiset materials and draw quality tiers from beacon-seeded pseudo-random streamsPhase 2 (Near-Term)
Atomic Trade Escrowdreggnet-trade/src/lib.rs (45ef8a7d2)Sealed escrow atomic swap: asset $\leftrightarrow$ asset or asset $\leftrightarrow$ $DREGGStalled during minidregg bake-off/Users/ember/dev/dclutch/ & starbridge-escrow-marketConnect DelveTalk2 item trading directly to DClutch escrow markets on SolanaPhase 2 (Near-Term)
Persistent Social Taverndreggnet-tavern/src/lib.rs (45ef8a7d2)Shared board + private stall cells; live SSE receipt streamShift of focus to Tulip Zulip bots/Users/ember/dev/delvetalk2/world/kinds/parlour.kind + allgameRe-home tavern GM as a Tulip resident in allgame; stream live room receipts to Delvetown chat channelsPhase 2 (Near-Term)
Escrow Guild Treasurydreggnet-guild/src/lib.rs (45ef8a7d2)Shared room capability; depositor-only escrow reclaimStalled during minidregg bake-off/Users/ember/dev/delvetalk2/world/objects/Policy.obendImplement guild membership as capability sets in Policy.obend; custody group treasuries in DClutchPhase 2 (Near-Term)
Faction Standing & Decaydreggnet-faction/src/lib.rs (45ef8a7d2)Monotonic reputation; rival ceiling drop; betrayal flagsStalled during minidregg bake-off/Users/ember/dev/delvetalk2/world/objects/Tide.obendDrive reputation counters in DelveTalk2 NPCs; gate region entry on signed faction commitmentsPhase 2 (Near-Term)
Cross-Cell Equipmentdreggnet-gear/src/lib.rs (928c0eb12)ObservedFieldEquals multicell ownership; class-locked talentsStalled during minidregg bake-off/Users/ember/dev/delvetalk2/world/objects/Thing.obendEquip items by referencing Thing.obend pointers; bind class-locked abilities in DelveTalk2 agent action tablesPhase 2 (Near-Term)
Fused Leveling Companiondreggnet-companion/src/lib.rs (928c0eb12)Asset note + multicell leveling cell (ObservedFieldEquals)Stalled during minidregg bake-off/Users/ember/dev/delvetalk2/world/kinds/bond.kindBind DelveTalk2 companions to DClutch creature assets; gate player buffs on companion level progressionPhase 2 (Near-Term)
DO-ONCE ViewNode Frontenddreggnet-surfaces/src/lib.rs (1ba6a4a1e)Offering::render $\to$ deos_view::ViewNodeStalled during minidregg bake-off/Users/ember/dev/delvetalk2/site/Render DelveTalk2 kind/object states to ViewNode IR; pipe to Web, Discord, and Telegram Mini AppsPhase 3 (Expansion)
.dungeon Text DSL Compilerdungeon-on-dregg/src/dsl/ (1dd8a566b)Biconditional translation validation: MISSING vs PHANTOMStalled during minidregg bake-off/Users/ember/dev/delvetalk2/world/kinds/room.kindRecompile .dungeon files directly into DelveTalk2 room.kind graph structuresPhase 1 (Immediate)
Attested Dungeon Masterattested-dm/src/lib.rs & GOAL-COLLECTIVE-FICTION-DEMO.mdslot_confinement theorem (Lean) + DmCaps::authorize gateLocal Ollama setup too heavy for CIallgame residents (claude_resident, kimi_resident)Host resident DMs in Allgame; route narrative effects through typed DelveTalk2 command channelsPhase 1 (Immediate)
Fail-Closed Spend Ceilingnarrator/src/lib.rs (aa8847c80)Hard $20 BudgetLedger with pre-flight reservation & file locksStalled during minidregg bake-offallgame LLM invocation gatewayWrap all external Claude/Bedrock API calls in allgame with BudgetLedger fail-closed reservationPhase 1 (Immediate)
Multiplayer Branch-Stitchmud-dregg/src/lib.rs (6f3309911)$(E, \le, \#)$ time-travel lattice; conflict refusalStalled during minidregg bake-off/Users/ember/dev/delvetalk2/world/Allow DelveTalk2 players to branch private world copies; reconcile state merges via settlement soundnessPhase 3 (Expansion)
Provably-Fair Generationprocgen-dregg/src/lib.rs (fd43cd6cd)Committed seed $\to$ DrawStream::draw_bounded $\to$ .dungeonStalled during minidregg bake-off/Users/ember/dev/delvetalk2/capsules/Seed daily dungeon worlds in DelveTalk2 using drand beacon entropy; verify byte-identical regenerationPhase 2 (Near-Term)
No-Cheat Leaderboardsugc-dregg/src/lib.rs (3b6776583)Re-execution of run transcripts against genesis world seedsStalled during minidregg bake-off/Users/ember/dev/delvetalk2/world/objects/Library.obendRank DelveTalk2 player completions on Library.obend by fresh deterministic journal replayPhase 2 (Near-Term)
ZK Hidden Hand Playdregg-multiway-tug/src/lib.rsPoseidon2 Merkle hand + Witnessed{MerkleMembership}Stalled during minidregg bake-off/Users/ember/dev/minidregg-wt/Integrate multiway-tug hidden-hand circuit into minidregg retained-mode proverPhase 3 (Expansion)
Path of Angels Mini-Gamespoa/ & metatheory/.../PathOfAngels/84 Lean modules; 1840 theorems; Signal, Relay, Salvage1-validator collapse; content-node drift~/dev/path-of-angels + DelveTalk2 station.kindPort Signal Triangulation, Relay Repair, and Salvage Lock to browser WASM over DelveTalk2 station roomsPhase 3 (Expansion)
Crowd-Stream Overlay Enginedocs/CROWD-STREAM-ENGINE-DESIGN.mdLive YouTube/Twitch Super Chat $\to$ CollectiveRound $\to$ OBS SSEStalled during minidregg bake-offDelveTalk2 web frontend + OBS overlayIngest YouTube/Twitch chats as weighted ballots into DelveTalk2 collective voting roundsPhase 3 (Expansion)
Game Design Solver Gatescripts/poa-design-gate.py (e42d699b9)Exhaustive transition table search: non-dominated verbs, strategic forksStalled during minidregg bake-offwhatember CI / DelveTalk2 authoring harnessRun design gate script across all generated DelveTalk2 rooms and .dungeon files before releasePhase 1 (Immediate)
Sealed-Bid Marketdreggnet-market/src/lib.rs (45ef8a7d2)Factory auction cells; sealed bids $\text{BLAKE3}(b \Vert v \Vert n)$; $\sum \delta = 0$Stalled during minidregg bake-off/Users/ember/dev/dclutch/Re-home uniform-price sealed-bid auction cycles to DClutch on-chain order booksPhase 2 (Near-Term)
Metered Compute Exchangedreggnet-compute/src/lib.rs (45ef8a7d2)Escrowed job budgets; FieldLteField(BID <= BUDGET); flashwell settleStalled during minidregg bake-off/Users/ember/dev/allgame/Use compute exchange to lease GPU/LLM inference time among Tulip residentsPhase 3 (Expansion)
Discord Identity Derivationdreggnet-discord-identity/src/lib.rs`BLAKE3_derive_key("dregg-discord-bot-v1", bot_secretuid)`Excluded bot workspaceDelveTalk2 Discord bridge
Confined Agent Offeringdreggnet-hermes/src/lib.rs (45ef8a7d2)AgentRuntime with ToolGrant rate caps & Charge value budgetsStalled during minidregg bake-offallgame resident harnessesWrap resident agent tool execution in ToolGateway metered turnsPhase 2 (Near-Term)
Sandboxed Grain Appsdreggnet-grain/src/lib.rs (45ef8a7d2)Sandstorm app grains with calls_made <= rate_limit rate legsStalled during minidregg bake-off/Users/ember/dev/delvetalk2/capsules/Host interactive mini-apps inside DelveTalk2 rooms via capability-sandboxed grainsPhase 3 (Expansion)
Governed Namespacesdreggnet-names/src/lib.rs (45ef8a7d2)Sovereign name cells; WriteOnce(NAME_HASH) first-claim toothStalled during minidregg bake-offDelveTalk2 identity layerResolve DelveTalk2 room and actor handles to verifiable on-chain sovereign name cellsPhase 3 (Expansion)
Collaborative Documentsdreggnet-doc/src/lib.rs (45ef8a7d2)Multi-editor document turns; antichain conflict refusalsStalled during minidregg bake-offDelveTalk2 lore editorEnable multi-user collaborative world/lore authoring inside DelveTalk2Phase 3 (Expansion)
Tournament Bracketsdreggnet-tournament/src/lib.rs (45ef8a7d2)Seeded elimination bracket; advancement gated strictly on verified winStalled during minidregg bake-offDelveTalk2 competitive arcadeRun automated speedrun and high-score tournaments over DelveTalk2 daily dungeonsPhase 3 (Expansion)
Deterministic Spritesdreggnet-sprite/src/lib.rs (1ba6a4a1e)Procedural seed-derived SVG generator for loot and charactersStalled during minidregg bake-offDelveTalk2 web frontendDeterministically render item and avatar SVGs directly from content addresses in-browserPhase 1 (Immediate)
Solana Holding Admissionpoa-solana-gate/src/lib.rs (7897c8eab)Token-2022 balance verification; one-use HoldingCapability noncesStalled during minidregg bake-off/Users/ember/dev/dclutch/Gate access to exclusive DelveTalk2 rooms and features by proving $DREGG holding via Solana RPCPhase 2 (Near-Term)
Dossier 19 · Breadstuffs Federation, Cross-Chain Light Clients, `pg-dregg`, `fhir` & Polisware Diaspora (50 KB)
Working field notes & source index below, with theorem anchors, code pointers, git history, and open questions.

Strand 19 — The Breadstuffs Diaspora: Federation, Multi-Chain Settlement, Blocklace/CapTP, Commons Governance, Polisware, and the Enterprise Verticals

Primary-source dossier. Compiled 2026-11-02 from the repositories under /Users/ember/dev/, their git logs, their in-tree goal documents, and the Claude memory stores under ~/.claude/projects/-Users-ember-dev-breadstuffs/memory/. Every commit SHA, Lean 4 / HOL4 / ACL2 theorem signature, Rust struct, and verbatim quotation below was extracted directly from the local repositories and memory records, not reconstructed from memory. Where an artifact was parked or a proof left uncompleted, the text notes the exact failure mode, missing lemma, or architectural friction. Scope boundary. /Users/ember/hellas is strictly third-party historical context (Ember's prior work on workload verification, which informed the genesis of Dregg after departure) and is not claimed as open property. All repos outside /Users/ember/dev/whatember/ are treated as strictly read-only.


0. Executive Map & Architectural Lineage

The monolithic repository /Users/ember/dev/breadstuffs was the expansive crucible of Ember's systems engineering. Between May and August 2026, breadstuffs housed simultaneously: a Byzantine fault-tolerant consensus engine over equivocation-free DAGs (blocklace), an object-capability network transport protocol (captp), a multi-chain non-custodial settlement and light-client matrix (chain, cosmos-*, solana-*, eth-lightclient, mina), an axiomatic democratic deliberation and collective governance substrate (collective-choice, commons-arbiter, metatheory/Polis), and an enterprise/database infrastructure tier (pg-dregg, fhir, durable-workflow).

Under the pressure of build times, multi-tenant agent contention, and proving overheads, this monolith fractured into a specialized diaspora. The table and ASCII lineage below trace where those ideas landed, what succeeded, and what remained parked.

text
                                    ┌────────────────── breadstuffs (monolith) ──────────────────┐
                                    │  consensus: blocklace · captp · dfa-federation · coord     │
                                    │  chains: cosmos-* · solana-* · eth-lightclient · mina      │
                                    │  governance: collective-choice · commons-arbiter · polis   │
                                    │  enterprise: pg-dregg · fhir · durable-workflow            │
                                    └──────────────────────────────┬─────────────────────────────┘
                                                                   │
         ┌─────────────────────┬───────────────────────────┼───────────────────────────┬──────────────────────┐
         ▼                     ▼                           ▼                           ▼                      ▼
┌──────────────────┐  ┌──────────────────┐  ┌─────────────────────────────┐  ┌───────────────────┐  ┌──────────────────┐
│ dragons-clutch   │  │        fn        │  │      svenvs ⋈ Polis         │  │     verekos       │  │ nanodregg/obend  │
│   → dclutch      │  │ BPv7 + NNTP      │  │ Formal non-domination       │  │ Claim/receipt     │  │ Simplex BFT      │
│ Solana SBF       │  │ Store-and-fwd    │  │ "Verify cage, not animal"   │  │ Mathlib-free      │  │ Obend.Consensus  │
│ 8 programs,      │  │ ACL2-verified    │  │ svenvs: HOL4 libertyScript  │  │ Paused 2026-10-19 │  │ Obend.Ledger     │
│ Cohorts 8 → 17   │  │ RFC 9171–9174    │  │ Polis: 66 Lean files        │  │ → minidregg       │  │ Pure Lean / Bend │
└──────────────────┘  └──────────────────┘  └─────────────────────────────┘  └───────────────────┘  └──────────────────┘

High-Level Subsystem Status Summary

SubsystemOriginal breadstuffs PathDiaspora DescendantMaturity StatusPrimary Bottleneck / Park Reason
Consensus / DAG/blocklace, /coordbreadstuffs (origin/main 18b3abebf), nanodregg, obendParallel Tracksblocklace remains active in breadstuffs (18b3abebf, Nov 14, 2026; OPEN-CM-XSORT open), while obend pursues Simplex BFT over Vault (Obend/Consensus).
Object Capabilities/captpfn, obend/lean/Obend/VaultParked / AbsorbedML-DSA PQ test harness aborts (SIGABRT); distributed GC cross-federation complexity.
Store & Forward/captp/src/store_forward.rsfn (specs/bp-path.md)Graduated (L4)Scaled into RFC 9171–9174 Bundle Protocol v7 with formal ACL2 proofs in fn.
Multi-Chain Settle/chain, /cosmos-*, /solana-*dclutch, obendL3 (Local / Devnet)25-lane Groth16 verified on EVM, Solana SBF, and CosmWasm; blocked on production MPC setup.
Semantic Lightclients/eth-lightclient, /minaIn-tree docs/L2 PrototypeMina Pasta curve-cycle mismatch; Ethereum sync committee lacked BLS12-381 in-tree.
Cross-Chain Gov/dregg-interchain-govobend/lean/Obend/LedgerParkedEdge tag agreement held only by mirrored constants; lacked proof-of-holding derivation.
Polisware/metatheory/Polis/svenvsL3 (Theoretic Apex)66 Lean files axiom-clean; real LLM agent runtime harness unbuilt; unified framing inflated.
AI Arbitration/commons-arbitero7vn (Oracle7)ParkedLexical prompt-injection filter ({{ detection); lacked semantic proof of non-domination.
Agent App Store/grain-commons, /hosted-*verekos, minidreggParkedRequired Sandstorm .spk package pipeline and lived under operated-layer infrastructure debts.
Postgres RLS Engine/pg-dreggminidreggParkedcargo-pgrx excluded from CI workspace; required managed PG18 runtime; Lean FFI glue churn.
FHE Product DSL/fhir (fhIR-0)obend/HardwareL1 SeedSix-part admissibility theorem remained in metatheory/ unproved; FHE matvec size envelopes.
Durable Workflow/durable-workflowhosted-durableL2 PrototypeC-int ABI mismatch on setrlimit (Linux-gnu); duroxide sqlite backend lock contention.

1. What Lived in breadstuffs (The Monolithic Peak)

1.1. Federation, Consensus, and the Blocklace

The federation layer was designed to execute multi-node consensus with verified finality. Its primary documents were GOAL-FEDERATION.md and memory file project-federation-payoff.md.

The n=4 Federation Payoff (2026-07-07)

On 2026-07-07 (cv session 5d12f365-a748-4e7a-b4ca-99a0291fef49), Ember achieved the core federation milestone: a foreign, non-operator client signed its own transfer, submitted via POST /turns/submit, and it stream-finalized cross-node on a verified-executor $n=4$ mesh across two physical machines (hbox at 192.168.50.39 and nextop at 192.168.50.130). The test harness node/tests/payoff_client_turn.rs verified:

  • Dest account funded across all 4 nodes;
  • Block heights advanced uniformly: [1,1,1,1] → [2,2,2,2];
  • Finalized in ~25 seconds over two clean rounds under DREGG_TEST_REQUIRE_FINALITY=1.

The $O(\text{history})$ Wall in Verified Finality

The path to this payoff uncovered a severe architectural defect. When running with the verified Lean tau-order finality gate (VerifiedFinality::compute_order, node/src/finality_gate.rs:150 calling dregg_tau_order FFI into Lean's tauOrderFast), an $n=3$ node mesh would plateau while $n=4$ progressed. The investigation proved this was not a correctness fault, but a catastrophic complexity cliff:

*"The verified-Lean tau-order finality gate was $O(\text{history})$, recomputed from scratch on EVERY finality poll: the serial executor rebuilt the whole wire and Lean rebuilt PastCache/RoundCache over the full lace on each call, then discarded them. As the DAG grew, per-poll cost outpaced block production and the consensus stalled."* — project-federation-payoff.md

The Blocklace: Cordial Dissemination and Eviction-by-Past

The consensus engine in /Users/ember/dev/breadstuffs/blocklace/ implemented a Byzantine-repelling Directed Acyclic Graph (DAG) based on Almeida & Shapiro (2024, arXiv:2402.08068) and Cordial Miners (Shapiro et al., 2022, arXiv:2205.09174). In cv session ad7d9679 (sub-agent of 27b28e60, 2026-08-08, recorded in docs/reference/READING-BLOCKLACE-2026-08-08.md), Ember dispatched a literature review that uncovered a fundamental divergence between theory and implementation:

  1. The Live-Gossip Mutation Defect: The implementation in blocklace_sync.rs stated that the tau participant set must be a deterministic function of committed state alone. Yet handle_push called constitution.auto_evict(proof) on gossip arrival, mutating the live participant set. If Node A saw both halves of an equivocation fork and Node B saw only one, Node A ran at $n-1$ while Node B ran at $n$. Consequently, the wave leader formula wave_leader = participants[wave % n] selected different leaders on different nodes, creating a silent fork. Furthermore, because eviction was not a block in the DAG, a reboot silently reinstated the equivocator.
  1. The Fix (Commits 89238335a and 0eccd772d, 2026-08-08/09): Ember corrected this by moving eviction off gossip-arrival mutation into exclusion-by-past (node(b) ∉ byz(⌊b⌋)), making eviction a structural predicate over a block's causal closure:
rust
   // Commit 89238335ad618998cc74daa77328651add0ddcb8
   // Author: ember arlynx <ember@lunar.town>, Date: Sat Aug 8 17:37:36 2026 -0400
   // "feat(consensus)!: exclusion-by-past — eviction was a live mutation keyed on gossip arrival;
   // both papers make it a predicate over the block's own closure"

Followed by commit 0eccd772d1348fb29b79658cf134f3fc18fcfe6d (Sun Aug 9 05:26:58 2026), adding 2,115 insertions across blocklace/src/constitution.rs and blocklace/src/finality.rs to enforce the Cordial Miners Alg. 1:5 two-tips-per-miner flood cap and CreatorTips::Pair evidence requirements.

  1. DFA-Federation: To prevent circular dependencies (cell -> dfa -> federation -> captp -> cell), /Users/ember/dev/breadstuffs/dfa-federation/ isolated route updates. FederationQcVerifier verified route transitions using a BLS ThresholdQC over old_commitment || new_commitment:
rust
   pub struct FederationQcVerifier {
       committee: FederationCommittee,
   }

   impl ThresholdVerifier for FederationQcVerifier {
       fn verify_threshold(&self, expected_old: [u8; 32], new_comm: [u8; 32], proof_data: &[u8]) -> Result<(), RouteUpdateError> {
           let qc: ThresholdQC = postcard::from_bytes(proof_data)
               .map_err(|e| RouteUpdateError::ThresholdVerificationFailed(e.to_string()))?;
           let mut msg = [0u8; 64];
           msg[0..32].copy_from_slice(&expected_old);
           msg[32..64].copy_from_slice(&new_comm);
           self.committee.verify(&qc, &msg)
               .map_err(|e| RouteUpdateError::ThresholdVerificationFailed(e.to_string()))
       }
   }

1.2. Distributed Object Capabilities: captp

The CapTP implementation in /Users/ember/dev/breadstuffs/captp/ provided capability transport across federations, combining the classic Agoric/E CapTP architecture with post-quantum cryptography.

Architecture and Invariants

Defined in captp/src/lib.rs:

  • Sturdy References (uri::DreggUri): Portable, durable capability URIs of the form dregg://<fed-id>/<cell-id>/<swiss-number> that survive transport disconnections.
rust
  #[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize)]
  pub struct DreggUri {
      pub federation_id: [u8; 32],
      pub cell_id: CellId,
      pub swiss_number: [u8; 32],
  }
  • Swiss Table (sturdy::SwissTable): Maps unguessable bearer secrets ([u8; 32]) to live capability descriptors. Possession of the swiss number is itself the authorization.
  • Three-Party Handoffs (handoff::HandoffCertificate): Unforgeable offline capability transfers signed by the introducer, verifiable by any third party without contacting the issuer:
rust
  pub struct HandoffCertificate {
      pub introducer_pubkey: [u8; 32],
      pub recipient_pubkey: [u8; 32],
      pub sturdy_ref: DreggUri,
      pub caveat: Pred,
      pub signature: Signature, // Ed25519 or ML-DSA-65
  }
  • Distributed GC (gc): Cross-federation wire reference counting and drop tracking.
  • Hybrid Post-Quantum Cryptography (dregg-pq): ML-DSA-65 (FIPS 204) hybrid capability-handoff signatures and X25519 + ML-KEM-768 hybrid transport sessions to resist "Harvest Now, Decrypt Later" (HNDL) attacks.
  • Store-and-Forward Relay (store_forward::EncryptedMessage): Offline delivery using X25519 ECDH $\to$ HKDF-SHA256 $\to$ ChaCha20-Poly1305 AEAD. Relay operators inspect only ciphertexts; tampering or forging triggers authenticated decryption failure.

The PQ Harness Pitfall

A critical test-suite bug plagued captp. In commits d2181916c and d78b8512b, Ember discovered that 130 tests across captp and related crates had never executed because dregg-pq aborted the process with process::abort() when the Lean-verified ML-DSA-65 cores were missing in test binaries, causing tests to exit prematurely with bare SIGABRTs.


1.3. Multi-Chain Settlement and Semantic Light Clients

The multi-chain vision in breadstuffs was defined by Ember's strict doctrine:

"You only need wraps/bridges when you lack proofs about the other chain's state transitions — just be a full client of all the chains." — project-chain-participation-epoch.md

The 25-Lane Proof Statement

Rather than trusting centralized bridge multisigs, Dregg’s settlement object was an IVC STARK fold compressed into a gnark Groth16 proof over BN254. The canonical statement was defined as 25 BabyBear field lanes ($p = 2013265921 = 0x78000001$):

text
Statement = genesis_root[8] ++ final_root[8] ++ num_turns[1] ++ chain_digest[8]

With SEG_ANCHOR_WIDTH = SEG_DIGEST_WIDTH = 8 (circuit-prove/src/ivc_turn_chain.rs:284).

Cross-Chain Settlement Parity

The 25-lane BN254 Groth16 proof was verified natively across three disparate runtime architectures:

  1. EVM (chain/contracts/DreggSettlement.sol): Concrete 25-lane contract using Ethereum's ecPairing precompile, paired with DreggGroth16Verifier25.sol, DreggVault.sol, and DreggCredentialGate.sol (commits 53efc5503, b997f9cef, ee9f9d887, 35af20966).
solidity
   contract DreggSettlement is IDreggSettlement {
       uint256 public constant BABYBEAR_P = 2013265921;
       IGroth16Verifier25 public immutable verifier;

       uint32[8] private _genesisLanes;
       uint32[8] private _provenLanes;
       uint64 private _provenHeight;

       function settle(
           uint32[8] calldata newRootLanes,
           uint64 numTurns,
           uint32[8] calldata chainDigestLanes,
           uint256[8] calldata proof
       ) external override {
           uint256[25] memory publicInputs;
           // pack _provenLanes, newRootLanes, numTurns, chainDigestLanes...
           require(verifier.verifyProof(proof, publicInputs), "invalid settlement proof");
           _provenLanes = newRootLanes;
           _provenHeight += numTurns;
           _provenRoots[keccak256(abi.encodePacked(newRootLanes))] = _provenHeight + 1;
       }
   }

The VK Commitment Flaw Fixed (Flag Day 2026-07-30): The original contract stored bytes32 private immutable _verifyingKeyHash which was never compared against the key used in ecPairing. It was corrected so that verifyingKeyHash() reads directly from verifier.vkDigest(), making mismatch structurally impossible.

  1. Cosmos (cosmos-settlement/src/lib.rs): A CosmWasm contract verifying the identical 25-lane proof and Pedersen commitment using ark-bn254, without requiring a Cosmos-native curve instantiation.
  1. Solana (solana-settlement/src/lib.rs): A native Solana SBF program (no Anchor) verifying the identical proof on-chain using Solana's alt_bn128 syscalls (solana_bn254::prelude).

Semantic Light Clients

  • EVM (eth-lightclient): Verified Ethereum sync-committee attestations, weak-subjectivity checkpoints, and EIP-1186 Merkle Patricia Trie (MPT) storage proofs. In commit 0aba72c65788 (2026-07-29), the sync committee rotation was transferred to Lean verification after finding that store::bootstrap_committee had bypassed the verified gate. Commit dba55c04b removed an accidental DREGG_LANE_AA_MUTANT=accept bypass.
  • Cosmos (cosmos-lightclient): Verified Tendermint $\ge 2/3$ commit signatures and ICS-23 IAVL Merkle tree membership.
  • Solana (dregg-bridge/src/solana_holdings.rs): Verified Tower-BFT finalized bank hashes and stake-weighted supermajorities.
  • Mina (GOAL-MINA-SEMANTIC-LIGHTCLIENTS.md): Reached acceptance of real Mina Devnet block 539508 via a Lean Kimchi/Pickles verifier (P0–P6), but the Pasta-vs-BabyBear curve-cycle mismatch prevented full arithmetic deployment.

Non-Custodial Cross-Chain Governance: dregg-interchain-gov

In /Users/ember/dev/breadstuffs/dregg-interchain-gov/src/lib.rs, Ember joined all three foreign light clients into a single compiled governance seam. Foreign token holders proved their balances on their native chain via read proofs:

  • Solana: ProvenHolding
  • EVM: ProvenErc20Holding
  • Cosmos: ProvenCosmosFact

The holder signed an owner -> voter binding (Ed25519 on Solana, EIP-191 on EVM, Cosmos ADR-036). The compile-time pins _EVM_TAG_PINNED and _COSMOS_TAG_PINNED locked the tag mapping (Solana = 0, EVM = 1, Cosmos = 2), allowing foreign balances to flow directly into Dregg's [CollectiveChoice] tally without locking, bridging, or wrapping a single token:

rust
pub fn proven_holding_to_governance(
    proven: &ProvenCosmosFact,
    chain_id: ChainId,
    nullifier_salt: [u8; 32],
) -> Result<ProvenForeignHolding, ForeignFieldsError> {
    let fields = cosmos_lightclient::foreign_holding_fields(proven, nullifier_salt)?;
    ProvenForeignHolding::from_foreign_fields(chain_id, fields)
}

1.4. Democratic Deliberation, Commons Governance, and Polisware

breadstuffs contained one of the most mathematically ambitious governance stacks ever designed, spanning game theory, Ostrom commons management, and formal social choice.

Collective Choice (collective-choice/src/lib.rs)

Composed four formal mechanisms into a unified VoteEngine:

  1. One-Vote-per-Ballot: Ballot cell with a WriteOnce(VOTE) constraint.
  2. Monotone Tally: Poll cells with Monotonic option tallies, preventing replay attacks from lowering counts.
  3. Quorum Gate: In-cell affine inequality constraint: $$\text{AffineLe} \{ M \cdot \text{RESOLVED} - \sum \text{TALLY}_i \le 0 \}$$
  4. Non-Amplifying Delegation (Liquid Democracy): Macaroon attenuation via Mandate::sub_delegate (DelegTree::no_amplify).
  5. One-Vote Nullifiers: Third layer of defense preventing double voting across the network.

The Commons Arbiter (commons-arbiter/src/lib.rs)

An "AI authority nobody owns" (docs/deos/UNCENSORABLE-UTILITY.md). An autonomous LLM evaluator governed by a community Rubric:

  • A member submits a Case (content + rubric ID).
  • An attested LLM session emits a ruling carrying a ZkOracleAttestation.
  • Anti-Prompt-Injection Proof: The attestation proves the ruling is authentic, well-formed JSON, and injection-free (proved free of {{ handlebars delimiters via a verified negative-regex matcher).
  • Quorum Ratification: A ruling becomes final only when a committee of $n$ independent operators ratifies it ($\lfloor 2n/3 \rfloor + 1$).

Grain Commons (grain-commons/src/lib.rs)

A decentralized app-store and execution commons for hosted AI agents (Sandstorm grains):

  • Signed .spk packages where the App ID is the author's Ed25519 public key.
  • Committed Umem memory heaps with tamper-proof Pedigree Merkle trees tracing every code fork.
  • GenesisAgent minting bounded child sub-agents with attenuated capability ceilings.

Polisware: The 66-File Lean Corpus (metatheory/Polis/)

Polisware was the formal mathematical constitution for multi-agent coexistence, synthesizing Ember's work on svenvs with computable politics. The spine is defined in metatheory/Polis/Polis.lean:

lean
/-- **`polis_safety` — verify the cage, not the animal.** For a sound policy, a safe
shield, and a safe start, the enveloped system keeps the floor at EVERY step, for EVERY
controller. The controller is universally quantified and never inspected: no psychometric
classification of the inhabitant can be load-bearing for the guarantee. -/
theorem polis_safety
    {step : State → Action → State} {safe : Floor State}
    {pol : Policy State Action} {shield : State → Action} {init : State}
    (sound : SoundPolicy step safe pol)
    (shieldSafe : ∀ s, safe s → safe (step s (shield s)))
    (initSafe : safe init) :
    ∀ (ctrl : State → Action) (n : Nat),
      safe (traj step (envAct pol shield ctrl) init n) := by
  intro ctrl
  have step1 : ∀ s, safe s → safe (step s (envAct pol shield ctrl s)) := by
    intro s hs
    unfold envAct
    by_cases hp : pol s (ctrl s)
    · rw [if_pos hp]; exact sound s (ctrl s) hs hp
    · rw [if_neg hp]; exact shieldSafe s hs
  intro n
  induction n with
  | zero => exact initSafe
  | succ k ih => exact step1 _ ih

Key constitutional theorems in Polis.lean:

  • Honesty Discrimination via Typing (lines 213–234):
lean
  inductive ClauseEvidence (c : Clause) : Type where
    | structuralProof (proof : c.claim) : ClauseEvidence c
    | adjudicationCost (clerkCost : Nat) : ClauseEvidence c
    | jurisdictionBoundary (reason : String) : ClauseEvidence c

  theorem structural_requires_proof {c : Clause} (e : ClauseEvidence c)
      (h : e.status = LawStatus.structural) : c.claim := by
    cases e with
    | structuralProof p => exact p
    | adjudicationCost _ => simp only [ClauseEvidence.status] at h; exact absurd h (by decide)
    | jurisdictionBoundary _ => simp only [ClauseEvidence.status] at h; exact absurd h (by decide)
  • Monotone Amendment and Non-Regression (lines 279–314):
lean
  theorem regressive_amendment_inert (frozenMin : Agent → Floor State)
      (am : Amendment Agent State) (C : Constitution Agent State)
      (h : ¬ NonRegressive frozenMin am C) :
      guardedApply frozenMin am C = C := by
    unfold guardedApply; rw [if_neg h]

  theorem amendment_stream_nonregression (frozenMin : Agent → Floor State)
      (ams : Nat → Amendment Agent State) {C : Constitution Agent State}
      (hwf : WellFormed frozenMin C) :
      ∀ n, WellFormed frozenMin (amendStream frozenMin ams C n) := by
    intro n
    induction n with
    | zero => exact hwf
    | succ k ih => exact guardedApply_wellFormed frozenMin (ams k) ih
  • Schism as Empty Meet (line 335):
lean
  theorem disjoint_floors_no_polis (floors : Agent → Floor State)
      (hempty : ¬ ∃ s, ∀ i, floors i s) : ¬ CanFormPolis floors := by
    rintro ⟨s, hs⟩
    exact hempty ⟨s, hs⟩

A polis exists only where exported floors share an inhabitant. If floors are disjoint, there is no common court; this is not enforced by censorship or majority vote, but is a structural derivation.


1.5. Enterprise, Storage, and Durable Database Verticals

pg-dregg: Verifiable Postgres Extension

In /Users/ember/dev/breadstuffs/pg-dregg/, Ember built a PostgreSQL extension (targeting PostgreSQL 18 via cargo-pgrx) that made Dregg's object-capability model the native Row-Level Security (RLS) engine for relational data:

  • src/authz.rs: Tier-A capability decoding, verification LRU, and caveat algebra.
  • src/jsonpath.rs: Compiled Dregg Pred predicates directly into SQL jsonpath filter expressions:
rust
  // Structural translation table from pg-dregg/src/jsonpath.rs
  // Pred::True                   => "true"
  // Pred::False                  => "false"
  // Pred::AttrEq{key, value}     => "@.key == \"value\""
  // Pred::AttrPrefix{key, pfx}   => "@.key starts with \"pfx\""
  // Pred::NotBefore{at}          => "@.clock >= at"
  // Pred::NotAfter{at}           => "@.clock <= at"
  // Pred::Within{nb, na}         => "@.clock >= nb && @.clock <= na"
  // Pred::AllOf([p...])          => "(f1 && f2 && ...)"
  // Pred::AnyOf([p...])          => "(f1 || f2 || ...)"
  // Pred::Not(p)                 => "!(f)"
  • src/mirror.rs: Tier-B/C row types with a RootChain anti-substitution mechanism.
  • src/attest.rs: Tier-C range-attestation proof gate linking SQL queries to zk-circuit receipts.
  • src/workflow.rs: In-database durable workflow execution.

fhir (fhIR-0): FHE Convex Optimization DSL

Despite its name, /Users/ember/dev/breadstuffs/fhir/ was not medical HL7 FHIR; it was fhIR-0: Fully Homomorphic Encryption Intermediate Representation:

  • Typed DSL for financial orders and products on top of the fhEgg / Cert-F solver.
  • Products carried a convex program shape (LP / QP / aggregation) and a privacy tier: Dark (Tier 0, FHE, public matrices only), Shielded (Tier 1, solver sees private matrices), and Open (Tier 2, public).
  • types.rs defined the visibility and admissibility rule:
rust
  #[derive(Clone, Copy, PartialEq, Eq, Debug, Serialize)]
  pub enum Visibility {
      Public,
      Private,
  }

  pub struct ProgramType {
      pub curvature: Curvature,
      pub matrices: Vec<Visibility>,
      pub cones: Vec<ConeKind>,
      pub dimension: usize,
  }
  • qp_certificate.rs: Generated exact SDD $\to$ PSD admission witnesses and strict KKT optimality witnesses.

durable-workflow, hosted-durable, and hosted-lease

A deterministic durable execution engine modeled after DBOS and Temporal:

  • durable-workflow/src/lib.rs: Workflows execute deterministic steps checkpointed to a duroxide store. Crashes resume at the first unfinished step without re-executing completed side effects.
  • hosted-durable/src/lib.rs: Conserving settlement rail charging per-period lease fees via recTransfer ($\sum \delta = 0$).
  • hosted-lease/src/lib.rs: Fused meters combining StandingObligation with prepaid_lease into a single atomic write.

auditable-fund and attested-dm

  • auditable-fund/src/lib.rs: An autonomous trading agent that cannot lie about fills. Every trade lands as a light-client verifiable receipt binding a zkOracle price attestation and an attested Hermes decision. A third party auditing the fund requires zero operator trust.
  • attested-dm/src/lib.rs: An attested AI Dungeon Master. Prompt integrity is certified via dregg_zkoracle_prove::render. Player input is strictly slot-confined and checked by a verified negative parser to ensure it cannot escape its data cell into system instructions.

dregg-storage-templates and dregg-stream-ingest

  • dregg-storage-templates/src/cap_inbox.rs: Standard cell templates for CapInbox (deposit accounting + ring root), ProgrammableQueue, PubSubTopic, and BlindedQueue (custom ZK verifier).
  • dregg-stream-ingest/src/lib.rs: Normalized live chat events (YouTube Super Chats, Twitch cheers, TikTok gifts) into paid, weighted ballots for collective voting.

2. The Diaspora: How Ideas and Subsystems Spread Out

When breadstuffs was archived, its core components migrated to specialized repositories across /Users/ember/dev/.

text
breadstuffs Subsystem                 Descendant Repository                   Key Transformation
─────────────────────────────────────────────────────────────────────────────────────────────────────────────
Solana Bridge & Settlement    ───▶   dragons-clutch ──▶ dclutch             Decomposed monolith into 8 SBF programs,
                                                                             2,646 Lean theorems, Cohort 17 on devnet.
Store-and-Forward CapTP       ───▶   fn                                      Transformed into formal RFC 9171–9174
                                                                             BPv7 bundle protocol with ACL2 proofs.
Claim & Receipt Ledgers       ───▶   verekos ──▶ minidregg                   Paused EKOS half (2026-10-19); carried
                                                                             claim/receipt model into minidregg.
Polisware Non-Domination      ───▶   svenvs                                  Formulated the svenvs key insight:
                                                                             "govern trace-shape, never motive."
Blocklace / Cordial Miners    ───▶   nanodregg ──▶ obend                     Discarded DAG complexity for Simplex BFT
                                                                             (`Obend.Consensus.Simplex`).

2.1. Solana Settlement: dragons-clutch $\to$ dclutch

The multi-chain Solana work in breadstuffs (solana-lock, solana-settlement, dregg-bridge) moved first into dragons-clutch (gen-1, a 2.15 MB monolith ELF) and was subsequently rewritten into dclutch (gen-2). As documented in Strand 12:

  • Replaced the monolithic WAVE runtime with 8 discrete SBF programs;
  • Proved 2,646 Lean theorems establishing balance conservation and clearing fairness;
  • Deployed Cohorts 8 through 17 on Solana devnet (clutch.dregg.pro), culminating in the first retired market on any chain on 2026-09-06 (AvKSizb7...).

2.2. Asynchronous Federation: fn (BPv7 & NNTP)

The store-and-forward capability relay from captp/src/store_forward.rs and the decentralized news/federation mechanics evolved into /Users/ember/dev/fn/ ("formal news"):

  • Built as an NNTP news server where ACL2 makes every decision;
  • Implemented the full Delay-Tolerant Networking (DTN) Bundle Protocol v7 (BPv7) per RFC 9171, RFC 9172 (BPSec), RFC 9173, and RFC 9174;
  • Namespaces in ACL2:
  • fn-bpp-: Primary bundle blocks (CRC-16/CRC-32C, dtn and ipn endpoint IDs, fragment fields);
  • fn-bpb-: Canonical bundle frame codec;
  • fn-bpn-: Autonomous BPv7 forwarding node and RFC 6.1.1 status-report engine;
  • fn-bpf- & fn-bpfr-: Resumable bundle fragmentation and reassembly over overlapping slices;
  • Keystones certified in ACL2: fn-bpp-decode-of-encode, fn-bpc-dec-yields-shape, and fn-bpp-accepted-input-is-canonical-by-construction.

2.3. Claim/Receipt Ledgers: verekos $\to$ minidregg

The verifiable audit and receipt tracking pioneered in auditable-fund and pg-dregg was systematized in /Users/ember/dev/verekos:

  • Built 190 Lean theorems modeling claim-and-receipt ledgers over an enforced, Mathlib-free WorldCore boundary.
  • Pinned minidregg theorems: pinHolds, transclusion_pinned, toCausalPreimage_injective, transclude_uncovered_refused.
  • On 2026-10-19 (commit 89238335a / 8972945), Ember paused the EKOS half (project/CLOSEOUT-2026-10-19.md) and merged the claim/receipt architecture directly into minidregg.

2.4. Polisware and the svenvs Realization

The mathematical link between metatheory/Polis and /Users/ember/dev/svenvs was captured in project-polis-svenvs-key-insight.md:

*"svenvs resolves the protected-remainder paradox FOR ENFORCEMENT, not for epistemology. The move: the inhabitant is the $\forall$-quantified term the proof never reasons about, so the shadow is real-but-INERT. And the floor is NEGATIVE and least-restrictive.* *The consequence that makes polisware possible: the same $\forall$-blindness that protects the remainder makes the politician UNGOVERNABLE-BY-INTERIOR. Therefore govern TRACE-SHAPE, not motive. That is not a compromise — it is the only coherent option, and it is exactly why polis_safety quantifies over an opaque controller that is never inspected. 'Verify the cage, not the animal.'* *dregg's two floors ARE svenvs's two theorems. The authority floor corresponds to safety_preservation; the human floor corresponds to corrigibility_floor_holds."*

In HOL4 (svenvs/liberty/libertyScript.sml), Ember proved that the cage is not gratuitously restrictive:

  • Theorem maxpol_sound: The maximal policy permitting actions that preserve safety is sound.
  • Theorem maxpol_is_greatest_sound: Every sound policy is a subset of maxpol.
  • Theorem envelope_is_least_restrictive: Any policy permitting an action forbidden by maxpol is provably unsound for some inhabitant.
  • Theorem envelope_never_overrides_safe_action: Safe actions pass through untouched.

2.5. Consensus Transformation: nanodregg $\to$ obend

The complexity of the blocklace DAG and the O(history) finality bottlenecks forced a dramatic simplification:

  1. nanodregg stripped consensus down to its essential state machine, implementing Simplex BFT.
  2. In /Users/ember/dev/obend/lean/Obend/Consensus/Simplex.lean, Ember unified this consensus lineage:
  3. Absorbed minidregg/Kernel/GenericSimplex.lean, GenericSimplexQuorum.lean, and GenericSimplexCertificateSafety.lean;
  4. Eliminated DAG traversal in favor of view-based notarization certificates (NotarizationCert, QuorumCert):
lean
     structure Proposal where
       view : View
       height : Nat
       parent : Block
       turn : Option Obend.Vault.VaultTurn
       deriving BEq

     structure Message where
       sender : NodeId
       view : View
       kind : Kind
       value : Argument
       deriving BEq
  • Bound block payloads directly to Option Obend.Vault.VaultTurn, proving constructive safety and liveness without Mathlib.

3. What Got Dropped, Parked, or Underelaborated

An archaeological excavation reveals why several brilliant systems in breadstuffs were left behind:

text
Dropped / Parked Component        Root Cause of Abandonment / Park
─────────────────────────────────────────────────────────────────────────────────────────────────────────────
pg-dregg                          pgrx cdylib clashed with workspace cargo check; required external PostgreSQL 18
                                  daemon; Lean-FFI bridge was fragile and unmaintained.
fhIR-0 (FHE Order DSL)            Admissibility theorem ("admissible iff it compiles") unproved; FHE public-matrix
                                  envelope too restrictive for general order-book clearing.
Polisware & Commons Arbiter       Real LLM agent runtime harness never deployed; prompt-injection defense relied on
                                  lexical `{{` parsing rather than semantic non-domination.
Auditable Fund / Attested DM      Strictly paper-trading; lacked live broker API integration; prompt integrity was
                                  proven at template-expansion time but output behavior remained unverified.
Blocklace + CapTP                 O(history) finality gate bottleneck; live-gossip eviction bugs; ML-DSA test harness
                                  SIGABRT crashes; multi-round DAG sync complexity.
Cosmos / EVM / Solana Settle      gnark BN254 Groth16 setups relied on single-party dev keys (toxic waste);
                                  lacked on-chain proof-of-holdings derivation to parameterize voting weight.
Mina / Kimchi Light Client        Curve-cycle mismatch (Pasta / Tweedle vs BabyBear); IPA polynomial commitment verifier
                                  exceeded flat gate bounds in production.

Detailed Forensics

  1. pg-dregg: The dream of a verifiable Postgres extension with native capability RLS was functionally sound, but operationally marooned. Because cargo-pgrx manages its own PostgreSQL toolchain, it had to be excluded from the root Cargo.toml workspace. Whenever core types changed in breadstuffs/types, pg-dregg was left uncompiled. Furthermore, its Lean link relied on an ad-hoc C FFI wrapper (lean_producer.rs) rather than a unified compiler.
  1. fhIR-0: While the type checker correctly categorized programs into Dark/Shielded/Open tiers, the underlying mathematics required all constraint matrices in the Dark tier to be public (due to the limitations of FHE matrix-vector multiplication). This severely restricted its usefulness for arbitrary trading products. When the proving lane prioritized Plonky3 recursion, FHE development was paused.
  1. Polisware and the Unification Defect: The meta-review in project-polis-unification-correction.md (2026-07-06) found that the celebrated "non-domination $\equiv$ light-client-unfoolability" grand unification was a shape identity with zero shared proof content. The anti-vacuity checks were checking accept ≠ True rather than ∃ c, accept c, and DECO rung 5 was an empty tautology ($P \to P$). The theory was frozen at tag cut-trace-remainder-0.1 and never wired to live LLM swarms.
  1. Multi-Chain Governance Proof-of-Holdings Gap: As recorded in project-chain-participation-epoch.md, dregg-interchain-gov could verify read proofs from Solana, EVM, and Cosmos, but it had no mechanism to derive voting weight from holdings. Every voting engine defaulted to a static, hardcoded [u8; 32] voter set with one-voter-one-vote.

4. The Grand Revival Roadmap: Unifying with Modern Kernels

Our current formal kernels—obend, minidregg, svenvs, and o7vn—provide the exact architectural foundation needed to revive these dropped gems cleanly, eliminating the ad-hoc FFI layers and fragility of the 2026 monolith.

text
Dropped Gem                   Modern Kernel Anchor              Revival Architecture
─────────────────────────────────────────────────────────────────────────────────────────────────────────────
pg-dregg                      minidregg + o7vn                  Implement as a SQL Foreign Data Wrapper (FDW)
(Verifiable Postgres RLS)                                       where row-security policies evaluate against minidregg's
                                                                Mathlib-free WorldCore theorems.
fhIR-0                        obend/Hardware + Leaf             Formulate convex program admissibility as Open Hypergraph
(FHE Convex Product DSL)                                        transformations and Futamura projections in Bend/Lean.
Polisware & Commons Arbiter   svenvs + o7vn                     Ground democratic deliberation in svenvs's HOL4 libertyScript
(Non-Domination Deliberation)                                   theorems; replace regex prompt filters with Oracle7 specialists.
CapTP over Simplex            obend/Consensus + Vault           Embed sturdy capabilities directly into Obend.Vault hyperedges,
(Distributed Capabilities)                                      using Simplex BFT notarizations instead of blocklaces.
Multi-Chain Interchain Gov    obend/Ledger/Seat                 Map cross-chain read proofs (EVM MPT, Tendermint, Solana)
(Non-Custodial Weighting)                                       directly to Obend Ledger Seats with dynamic token-weighting.
Durable Workflow Engine       hosted-durable + minidregg        Express workflow checkpoint replays as deterministic
(Metered Orchestration)                                         minidregg Turn evaluations, eliminating duroxide SQLite locks.

4.1. Reviving pg-dregg via minidregg and o7vn

In breadstuffs, pg-dregg suffered from an unmaintained C-FFI link to Lean. In minidregg, the entire state transition system is self-contained, Mathlib-free, and compiles to clean C via lake build.

  • Revival Path: Compile minidregg/Kernel/World.lean directly into a PostgreSQL C extension.
  • Instead of maintaining complex LRU caches in Rust, PostgreSQL triggers execute World.step directly, checking RLS policies as pure Lean predicates.
  • o7vn's SQL reasoning specialists can synthesize and verify complex JSONPath queries with formal correctness proofs.

4.2. Reviving fhIR-0 via obend Open Hypergraphs

obend already models categorical computation through Obend.Hardware.OpenHypergraph and Futamura partial evaluation in Obend.Leaf.Futamura.

  • Revival Path: Define convex optimization cones as monoidal categories within OpenHypergraph.
  • The six-part admissibility theorem can be proved by showing that partial evaluation of a Dark-tier product eliminates all private matrix parameters, generating a pure public-matvec residual.
  • This elevates fhIR-0 from a speculative type checker to a verified program-extraction compiler for homomorphic clearing.

4.3. Reviving Polisware and Commons Arbiter via svenvs and o7vn

svenvs holds the ultimate formal formulation of non-domination in higher-order logic (libertyScript.sml).

  • Revival Path: Port Metatheory.Polis.polis_safety directly to the svenvs kernel.
  • Replace the crude {{ prompt-injection detection in commons-arbiter with semantic evaluation through o7vn's 145 verified native specialists.
  • When an AI arbiter rules on a dispute, it emits an o7vn reasoning certificate proving that its conclusion follows strictly from the constitutional rubric without hallucination or premise alteration.

4.4. Reviving CapTP over Simplex BFT in obend

The fatal flaw of CapTP in breadstuffs was its binding to the blocklace DAG, which incurred $O(\text{history})$ tau-ordering costs.

  • Revival Path: Port CapTP's sturdy references and 3-party handoffs to Obend.Consensus.Simplex.
  • Instead of reconciling DAG forks, capability handoffs land as linear transactions in Obend.Vault.VaultTurn.
  • Simplex notarization certificates (NotarizationCert) provide instant 2-step finality, allowing distributed object capabilities to pipeline promises across nodes with zero DAG latency.

4.5. Reviving Multi-Chain Non-Custodial Governance in obend/Ledger

dregg-interchain-gov proved that read proofs could verify cross-chain balances without wrapping tokens, but stalled on the proof-of-holdings weighting bridge.

  • Revival Path: Integrate foreign chain read proofs directly into Obend.Ledger.Seat.
  • An incoming EIP-1186 MPT proof (Ethereum) or ICS-23 proof (Cosmos) is verified against the finalized header in a SeatTurn.
  • The verified balance dynamically parameters the voter's voting power in Obend.Ledger.Book, finally realizing the vision of true cross-chain non-custodial governance.

Structured Revival Roadmap Table

PrimitiveOriginal Path in breadstuffsPrimary Dropped / Parked FlawModern Host KernelTarget Architecture & File SeamTarget Assurance Tier
pg-dreggpg-dregg/pgrx excluded from workspace; fragile C FFI glue.minidregg + o7vnLean-compiled Postgres FDW evaluating World.step directly.Tier 4 (Machine-Checked)
fhIR-0fhir/Admissibility theorem unproved; FHE size bounds.obend/HardwareMonoidal Open Hypergraphs + Futamura partial projection.Tier 3 (Sound Semantics)
Poliswaremetatheory/Polis/Anti-vacuity gaps; disconnected from live agents.svenvsPort polis_safety to HOL4; govern traces via libertyScript.Tier 4 (Machine-Checked)
commons-arbitercommons-arbiter/Lexical {{ injection checks; lacked semantic proof.o7vnAI rulings emit o7vn native reasoning certificates.Tier 3 (Certified Reasoning)
CapTPcaptp/Tied to $O(\text{history})$ blocklace DAG; PQ aborts.obend/ConsensusSturdy refs and 3-party handoffs over Simplex + Vault.Tier 4 (Linear Logic QC)
interchain-govdregg-interchain-gov/Lacked dynamic token-weighted balance derivation.obend/LedgerEIP-1186/ICS-23 read proofs parameterize Obend.Ledger.Seat.Tier 4 (Proof-Carrying)
durable-workflowdurable-workflow/SQLite duroxide lock contention; C-int ABI bugs.hosted-durableCheckpoint replays evaluated as deterministic Turns.Tier 3 (State-Machine Safe)
grain-commonsgrain-commons/Operated-layer Sandstorm .spk package debts.verekosClaim/receipt transclusion graphs replace .spk files.Tier 3 (Receipted Pedigree)

5. Primary Source Index

Repositories and Git Commits

  • /Users/ember/dev/breadstuffs:
  • 89238335ad61 (2026-08-08): Exclusion-by-past over blocklace DAG.
  • 0eccd772d134 (2026-08-09): Consensus literature pass (exclusion, catch-up, round gate).
  • 0aba72c65788 (2026-07-29): Ethereum sync committee rotation moved to Lean.
  • dba55c04b312 (2026-07-29): Removal of accidental DREGG_LANE_AA_MUTANT bypass.
  • 53efc5503, b997f9cef, ee9f9d887, 35af20966 (2026-07-11): 25-lane EVM settlement contract suite.
  • d2181916c, d78b8512b (2026-07-24): Fix for dregg-pq SIGABRT test suite aborts.
  • /Users/ember/dev/verekos:
  • 89729458b9ad (2026-10-19): Closeout of EKOS half; migration of claim/receipt model to minidregg.
  • /Users/ember/dev/obend:
  • 5f17752 (2026-10-31): Wave 3 architecture; Simplex consensus and multi-cell vault integration.
  • /Users/ember/dev/fn:
  • 43e6136d6, 05c318793: BPv7 bundle protocol implementation and lock discipline verification.

Goal and Architecture Documents

  • /Users/ember/dev/breadstuffs/GOAL-FEDERATION.md: Living federation execution goal and $n=4$ milestone.
  • /Users/ember/dev/breadstuffs/GOAL-MULTICHAIN-SETTLEMENT.md: 25-lane multi-chain proof settlement and wrap architecture.
  • /Users/ember/dev/breadstuffs/GOAL-MINA-SEMANTIC-LIGHTCLIENTS.md: Bidirectional semantic light client requirements.
  • /Users/ember/dev/breadstuffs/docs/reference/READING-BLOCKLACE-2026-08-08.md: Foundational literature review of Cordial Miners.

Claude Memory Files

  • project-federation-payoff.md: Detailed root-cause analysis of the $O(\text{history})$ finality gate bug.
  • project-polis-lean-corpus.md: Complete census of the 66 Polis Lean files.
  • project-polis-svenvs-key-insight.md: Derivation of trace-shape governance from svenvs $\forall$-blindness.
  • project-polis-unification-correction.md: Adversarial meta-review of Polis claims and anti-vacuity corrections.
  • project-chain-participation-epoch.md: Architectural specification of the 25-lane cross-chain settlement format.
  • project-lightclient-stark-true-peers.md: Analysis of folding verification gates across foreign chains.

Agent Transcripts (cv)

  • 5d12f365: Achieved $n=4$ cross-node stream finality payoff (2026-07-07).
  • a56148a1: Formalization of the 66 Polis Lean files and svenvs connection (2026-07-04/06).
  • 28920d6e: Launch of the multi-chain participation campaign and 25-lane census (2026-07-11).
  • 2ed94bdd: Mina Kimchi/Pickles Lean verifier development on Devnet block 539508 (2026-07-26/28).
  • ad7d9679: Sub-agent reader-blocklace literature pass on Almeida & Shapiro (2026-08-08).
Dossier 22 · Objective Bend (`obend`) Language, LTUO (`fareoo`), Equi-Recursive Bisimulation, Astra Effect Laws & 11 Spine Modules (86 KB)
Working field notes & source index below, with theorem anchors, code pointers, git history, and open questions.

Dossier 22: Objective Bend (obend) — Language Thesis, Equi-Recursive Core4 Type Theory, LTUO Grounding (fareoo), and the Astra Effect-Law & Consensus Document Substrate

[!IMPORTANT] Scope & Primary Source Grounding Whereas Dossier 07 examined obend's silicon/FPGA synthesis and dynamic partial reconfiguration leaf (Obend.Silicon, ExprCAM, ICAP, Sky130 DRC, CDNA3/FP4 equivalence), this monograph documents Objective Bend (obend) as a programming language, object/extension calculus, equi-recursive / algebraic-subtyping type theory, effect-law runner, and consensus document substrate. Every claim is grounded directly in three primary-source repositories on disk: 1. /Users/ember/dev/obend (Nov 1, 2026 snapshot: 386 commits through 7c699ca, 452 Lean 4 modules; 2026-11-14 Wave C3 Ground Truth (16fed219): 1,329 commits on dev, 597 Tier-1 green modules / 837 .lean files, 0 sorryAx, StageBInstances.lean sealed with 0 *Obligations, 256-bucket Vault/WorldTree.lean, Hardware/Device/* + 4 Stage-2C worktrees, and orchestrated by obebuilder obb v6/v7 aab42d2 — note that early sandstorm/ and obend-node scaffolding below was retired during the Wave C1/C2 cathedral convergence; see Dossier 28 §2). 2. /Users/ember/dev/obend-astra-decisions (frozen decision & lane archive at ed2878d / checkpoint-20261101 6266457, capturing 300 Lean modules, 35,127 kernel constants, 0 sorryAx, 59 LANE-*-BRIEF-*.md / REPORT-*.md files, and the October 29 – November 1, 2026 architectural debates). 3. /Users/ember/dev/fareoo (22 commits, Oct 1–3, 2026, HEAD d7ddf53: "Mirror current LTUO and ground stronger Objective Bend semantics").


1. Executive Thesis & Intellectual Genealogy (fareoo → minidregg + nanodregg → obend)

1.1 The Core Thesis of Objective Bend

Objective Bend opens README.md and THESIS.md with a single unifying sentence:

"Objects anyone can extend without forking, laws judged on every write, documents that remember why they say what they say." — /Users/ember/dev/obend/README.md, /Users/ember/dev/obend/THESIS.md

The architectural problem Objective Bend solves is stated in THESIS.md and formalized in paper/objective-bend.md:

  1. Software, prose, and machine-checked proofs rot by the same mechanism: an author publishes an artifact $A$; a downstream stranger wants $A$ except for one field, one policy, or one theorem assumption. Without open recursion and first-class specifications, the stranger must fork $A$—copying its source and severing its history—or beg the upstream maintainer to merge a patch.
  2. Simultaneously, references rot silently: when $A$ is revised, retired, or splits into two incompatible branches, downstream dependents either silently drift to unreviewed code (moving references like latest) or silently point to dead assumptions (pinned hashes whose underlying facts no longer hold).
  3. Why not plain Bend? As documented in ASTRA-SYNTHESIS-20261029.md, HigherCo's Bend / BendTT is an interaction-net functional language whose core discipline is source-level affinity: variables are used at most once so match frees graph nodes in place and fork-join parallelism needs no garbage collector or lock. However, Cook/Bracha/Faré open recursion (mix l u := λself. λsuper. u self (l self super) and fix s b := let rec self = s self (b self) in self) uses self twice in every composition step! Under a strict affine discipline, mix fails live-variable linearity whenever self carries captured methods or non-shareable state.
  4. Objective Bend's Synthesis: Objective Bend places lazy call-by-name open recursion (Core4), first-class specification/prototype reflection, C4 multiple-inheritance linearization, six-occasion state laws (Law), and contextual truth-maintenance slots (Slot, combining Nanevski–Pfenning–Pientka Contextual Modal Type Theory with Doyle's 1979 ATMS supports) at the top level of the language, while compiling closed leaf computations down to affine BendTT / interaction-net Book definitions via Obend.Leaf (lean/Obend/Leaf.lean) and Obend.Lower (lean/Obend/Lower.lean).
flowchart TD
    subgraph Lineage["Intellectual & Repository Genealogy (Oct–Nov 2026)"]
        LTUO["fareoo (d7ddf53, Oct 1–3, 2026)<br/>Faré LTUO Mirror (4.26 MB) + FAREOO.md + AUDIT.md"]
        MINI["minidregg (092027a2)<br/>Core4 CBN Machine, Spec Parser/C4, Six-Occasion Laws"]
        NANO["nanodregg (d81faee)<br/>Slot/Protocol, Fold Admission, Judge/LRAT, Receipt, Doc"]
        OBEND["obend (386 commits → 7c699ca)<br/>Unified 11-Module Spine + Equi-Recursive Core4 + Astra Lanes"]
        ASTRA["obend-astra-decisions (6266457 / ed2878d)<br/>Oct 29 – Nov 1 Decision Archive, 59 Lane Reports & Council Audits"]
    end

    LTUO -->|"Exposes K-LAW-INHERIT, K-KIND-CELL, K-CARRY<br/>& C4 Ordered-Presentation Bug"| MINI
    MINI -->|"Ports Core/ + Spec/ + Law/"| OBEND
    NANO -->|"Ports Slot/ + Fold/ + Judge/ + Receipt/ + Doc/ + Run/"| OBEND
    OBEND <-->|"Checkpoints & Evaluates 59 Formal Lanes"| ASTRA

1.2 The fareoo Excavation (d7ddf53): Auditing minidregg Against Faré's Lambda, the Ultimate Object

Between October 1 and October 3, 2026 (22 commits, from initial commit 8f1d452 to HEAD d7ddf53 "Mirror current LTUO and ground stronger Objective Bend semantics"), Ember created /Users/ember/dev/fareoo to mirror François-René Rideau's (Faré's) monograph Lambda, the Ultimate Object — Prototype- and Class- orientation, Typed Open Recursion, Purity, Hygiene, Winning Expressiveness vs Performance Trade-offs, Compatibility, Evolution, Great and Small, Using Metaphors, Category Theory, etc. (sources/ltuo-20261001.html, 4,266,596 bytes, SHA-256 4c1845238e04ddba74f13bfaa11dc6a863f54df0cf4532dd6c9059fa3e24e355, with the Objective Bend excerpt in sources/ltuo-objective-bend-20261003.txt).

In /Users/ember/dev/fareoo/FAREOO.md (1,003 lines) and /Users/ember/dev/fareoo/AUDIT.md (792 lines), Ember conducted a brutal clause-by-clause audit of minidregg's existing Objective Bend prototype against LTUO's twelve canonical object-orientation requirements:

  1. The Wegner (1987) Diagnosis: AUDIT.md concluded that minidregg's runtime was object-based, not object-oriented in Wegner's sense ($\text{OO} = \text{objects} + \text{classes/kinds} + \text{inheritance}$), because objects were flat mutable hash cells whose laws and kinds lived outside the open-recursive mix/fix tower.
  2. Three Fatal Structural Gaps Identified in AUDIT.md:
  3. K-LAW-INHERIT (Laws Replaced Whole — Below Teitelman's 1966 ADVISE Floor): In minidregg, adopting or extending a cell's law replaced the prior law outright (newLaw) rather than conjoining it along the C4 ancestry DAG (superLaw && selfLaw). AUDIT.md noted this fell below Warren Teitelman's 1966 PILOT / Interlisp ADVISE (BEFORE/AFTER/AROUND advice combinators) and Keene's 1989 CLOS :before/:after/:around method combination.
  4. K-KIND-CELL (Kinds as Second-Class Host Rebuilds): In minidregg, a "kind" was a host-level table entry rather than a first-class cell in the World with its own content-addressed pin, law, and extends lineage. Consequently, creating an instance (create --from K) or deriving a subkind (derive-kind --from K) could not be tracked by truth-maintenance supports.
  5. K-CARRY (Upgrade by Re-Genesis): Upgrading an object schema in minidregg required re-genesis or out-of-band host migration rather than in-place carry over Self/Super specifications.
  6. Ember's Formal Correction to Faré's LTUO §7.3.5 (The C4 Poset Invariance Bug): In LTUO §7.3.5 (sources/ltuo-objective-bend-20261003.txt), Faré claimed that the C4 multiple-inheritance linearization algorithm (which refines Barrett et al.'s 1996 Dylan C3 linearization by dropping the ad-hoc caller parent-list tie-breaker in favor of a suffix-closed hierarchy) is a canonical function of the bare inheritance poset/DAG. While proving C4's metatheory in Lean 4, Ember discovered a 2-parent counterexample (FAREOO.md, paper/objective-bend.md):
  7. Consider diamond inheritance where C inherits from two incomparable parents A and B (both over root O).
  8. As a bare poset, $\{C < A < O, C < B < O\}$ is symmetric between $A$ and $B$.
  9. Yet any total linearization must order either $A \prec B$ ([C, A, B, O] from presentation ((C A) (C B))) or $B \prec A$ ([C, B, A, O] from presentation ((C B) (C A))).
  10. Therefore, C4 is not invariant on the bare poset; it is invariant up to injective renaming of the ordered presentation (OrderedPresentationInvariant in lean/Obend/Spec/C4Vectors.lean, mechanically verified across 4,000 seeded DAGs and proved structurally in lean/Obend/Spec/Proofs/C4.lean).

1.3 Merging minidregg and nanodregg into obend

As recorded in GROUNDING.md and PLAN.md, obend was created by fusing two complementary lineages into a single zero-axiom (0 sorryAx) Lean 4 + Rust repository:

  • From minidregg (092027a2): The Core4 lazy call-by-name operational semantics (Obend.Core.Term, Obend.Core.Machine, Obend.Core.Soundness), the .obend surface parser and modular template elaborator (Obend.Spec), C4 linearization (Obend.Spec.C4), and the six-occasion law compiler (Obend.Law).
  • From nanodregg (d81faee): The open protocol & truth-maintenance slot calculus (Obend.Slot), the single-pass audited fold (Obend.Fold), the graded certificate judge and verified LRAT SAT checker (Obend.Judge), Merkle-backed receipts with epistemic disclosure dials (Obend.Receipt), pushout patch documents (Obend.Doc), and the guarded activity runner (Obend.Run).

2. The Core4 Calculus, Equi-Recursive Type Theory, and Algebraic Subtyping

2.1 Syntax of Core4 (Obend.Core.Term)

Unlike standard $\lambda$-calculi where objects are encoded via heavy Church/Bruce-Cardelli-Pierce existential packings, Obend.Core.Term (lean/Obend/Core/Term.lean) makes open recursion, specifications, prototypes, reflection, and yielded plans primitive constructors of a 26-constructor de Bruijn syntax:

lean
-- /Users/ember/dev/obend/lean/Obend/Core/Term.lean:17-58
inductive Primitive where
  | add | multiply | equal | conjunction | labelEqual
  | subtract | divide | less | lessEqual | modulo
  deriving Repr, DecidableEq

inductive Term where
  | hole (n : Nat) (σ : List Term)
  | bound (index : Nat)
  | lam (body : Term)
  | app (function argument : Term)
  | mix (inheritedExtension wrappingExtension : Term)
  | fix (specification inherited : Term)
  | specification (metadata extension : Term)
  | prototype (specification target : Term)
  | reflect (prototype : Term)
  | metadata (specification : Term)
  | project (prototype : Term)
  | nat (value : Nat)
  | boolean (value : Bool)
  | label (value : String)
  | binary (primitive : Primitive) (left right : Term)
  | extend (inherited : Term) (fields : List (String × Term))
  | record (fields : List (String × Term))
  | get (target : Term) (name : String)
  | ifZero (value zero successorBody : Term)
  | inject (label : String) (payload : Term)
  | case (scrutinee : Term) (arms : List (String × Term))
  | ifBool (condition whenTrue whenFalse : Term)
  | perform (plan : Term)
  | done (value : Term)
  | split (target : Term) (names : List String) (function : Term)
  deriving Repr

The object-oriented and contextual primitives in Term have exact operational reductions in lean/Obend/Core/Term.lean and lean/Obend/Core/Machine.lean:

  • specification m e and metadata s: Packages a metadata record $m$ (such as C4 lineage, law descriptors, or documentation) alongside an open-recursive extension function $e : \text{Self} \to \text{Super} \to \text{Row}$. metadata (specification m e) reduces in one step to $m$; applying (specification m e) as a function reduces to e.
  • mix l u (Wrapper/Mixin Composition): Given lower extension $l$ and upper wrapping extension $u$, mix l u is a weak-head value that, when applied to self and super, feeds l self super as the super argument of u self: $$\mathsf{mix}\;l\;u\;\mathsf{self}\;\mathsf{super} \;\longrightarrow\; u\;\mathsf{self}\;(l\;\mathsf{self}\;\mathsf{super})$$
  • fix s b (Lazy Open-Recursive Tie): Closes specification $s$ over base $b$ by passing the fixed point fix s b itself as self and b as super: $$\mathsf{fix}\;s\;b \;\longrightarrow\; s\;(\mathsf{fix}\;s\;b)\;b$$
  • prototype s t, project p, and reflect p: prototype s t pairs a live target $t$ with the specification $s$ that generated it without losing the open specification. project (prototype s t) ⟶ t extracts the live object (and get (prototype s t) f ⟶ get t f delegates field lookup directly), while reflect (prototype s t) ⟶ s recovers the unclosed specification $s$ so any stranger can later run fix (mix (reflect p) delta) base!
  • hole n σ (Contextual Modal Delayed Substitution): Added in commit d8db9f8 (astra/HOLE-REPAIR-SET-20261030.md), hole n σ represents contextual metavariable ?n suspended with an explicit delayed substitution $\sigma : \text{List Term}$ of its local de Bruijn context $\Delta_n$. Under Term.rename and Term.substitute, operations map functorially across $\sigma$: $$\text{substitute}\;\tau\;(\mathsf{hole}\;n\;\sigma) = \mathsf{hole}\;n\;(\sigma.\text{map}\;(\text{substitute}\;\tau))$$ When the hole is filled by a closed-in-$\Delta_n$ candidate $u$ via fillHole, $\sigma$ is discharged by applySub σ u (Obend.Core.Hole, lean/Obend/Core/Hole.lean), satisfying Contextual Modal Type Theory's (CMTT) commute law: $$\text{fillHole}\;n\;u\;(\text{substitute}\;\tau\;t) = \text{substitute}\;\tau\;(\text{fillHole}\;n\;u\;t)$$
  • perform plan and done value (Yielded Activities): perform p is not a value and has no internal reduction step; when the demand machine reaches perform p at top level, it halts in Yields p so the guarded runner (Obend.Run) can judge the emitted Plan against world laws and resume.
  • split target names function (Multiplicative Linear Record Destructor): Added in commit ef8f814 (lean/docs/REPORT-LINEAR-20261101.md), split forces target to a .record fields, extracts every field in names simultaneously, and passes them in order to curried function, enabling tensor-elimination ($\otimes$-elim) of records containing multiple linear capabilities without dropping siblings.

2.2 The Core4 Type & Quantity System (Obend.Core.Ty and Obend.Core.Check)

Core4's type grammar (lean/Obend/Core/Ty.lean) integrates a 4-point usage semiring (Quantity), closure reuse (Reuse), row-polymorphic records and variants with rigid Self/Super variables, first-class specification/prototype types, linear custody tokens, and computation (activity) types:

lean
-- /Users/ember/dev/obend/lean/Obend/Core/Ty.lean:7-35
inductive Quantity where
  | erased | affine | linear | unrestricted
  deriving Repr, DecidableEq
inductive Reuse where
  | once | reusable
  deriving Repr, DecidableEq

inductive Ty where
  | natural | label | boolean
  | variable (index : Nat)
  | arrow (reuse : Reuse) (parameter : Quantity) (domain codomain : Ty)
  | emptyRow
  | field (name : String) (member tail : Ty)
  | specification (metadata extension : Ty)
  | prototype (specification target : Ty)
  | custody (identity : Nat)
  | variant (row : Ty)
  | computation (plan response result : Ty)
  deriving Repr, @[expose] DecidableEq

Resourceful Shadow Retention (insertCanonical)

A subtle soundness hazard in row-typed languages with linearity or capability custody is resource hiding via field shadowing: if a record { cap : Custody(0), cap : Nat } overwrites cap with a shareable Nat, naive row canonicalization drops the shadowed Custody(0) field and makes the resulting record look shareable = true! Core4 prevents this in Ty.insertCanonical (lean/Obend/Core/Ty.lean):

lean
-- /Users/ember/dev/obend/lean/Obend/Core/Ty.lean:118-124
@[expose] def Ty.insertCanonical (row : Ty) (name : String) (member : Ty) : Ty :=
  match row with
  | .field prior old tail =>
      if name = prior then (if old.shareable then .field name member tail else .field name member row)
      else if name < prior then .field name member row
      else .field prior old (tail.insertCanonical name member)
  | tail => .field name member tail

When old.shareable is false (e.g., old is .custody id, a .once closure, an unshared rigid .variable, or a .computation), insertCanonical keeps row (old and its tail) behind the new .field name member, proving Ty.shareableUnder_canonical so canonical row equality can never launder a linear or once-only capability.

Coeffect Usage Vectors and linear Exact Discharge

In lean/Obend/Core/Check.lean, typing judgments synthesize a bottom-up usage vector uses : List Nat indexed by the lexical context $\Gamma$:

  • erased: count = 0 (safeQuantity .erased c := c == 0).
  • affine: count ≤ 1 (safeQuantity .affine c := c ≤ 1).
  • linear: count = 1 at binder discharge (discharged .linear c := c == 1), count ≤ 1 in subterms (safeQuantity .linear c := c ≤ 1), exact equality across control-flow joins (Quantity.agrees .linear c₁ c₂ := c₁ == c₂), and linearFree on lazy thunks that might discard or duplicate their environment (such as unforced arguments to non-.linear parameters, specification metadata discarded on application, or single-field .get projections that drop sibling fields).
  • unrestricted: any count : Nat, provided the argument type satisfies shareableUnder assumptions.shareableVariables and all captured variables in a .reusable closure are shareable (capturesNotShareable).

In lean/Obend/Core/LinearSoundness.lean (closed in commit ef8f814, REPORT-LINEAR-20261101.md), Ember proved that on the strict value-substitution operational semantics (ValueStep) where .linear arrows step by call-by-value (app_linear_val / app_linear_step), a single linear hypothesis at de Bruijn index 0 satisfies Linear1:

  • valueStep_preserves_typing_and_linear1: Every value step preserves both PartialTyping and the exact syntactic occurrence count (occ 0 = 1), and at a final value occ 0 = 0 (value_consumes_linear: the linear resource has been consumed exactly once).
  • Furthermore, LinearSoundness.lean proves the negative theorem cbn_breaks_linear_occ: under Core4's lazy call-by-name reduction, $(\lambda x :_1 A.\; x)\;((\lambda y :_1 A.\; \mathsf{nat}\;0)\;z)$ would step to $(\lambda y :_1 A.\; \mathsf{nat}\;0)\;z$ without forcing the inner redex first, justifying why split and strict linear beta are required for runtime resource accounting.

2.3 The Equi-Recursive Type Migration (89bd283, Oct 31, 2026)

One of the deepest type-theoretic achievements in obend is the equi-recursive type agreement replacement executed on October 31, 2026 (commit 89bd283, documented in lean/docs/REPORT-EQUIREC-20261031.md and the 131-entry mechanical journal lean/docs/EQUIREC-JOURNAL-20261031.txt).

Why 1-Step legacySameType Failed on Self-Referential Specifications

Prior to 89bd283, Obend.Core.Check.sameType only unfolded declared rigid type bounds assumptions.bounds by one step (or along acyclic alias chains). However, in LTUO / Objective Bend, any object whose method returns Self (such as a counter whose inc : Nat → Self returns a new counter, or a stream whose tail : Self returns the rest of the stream) induces a cyclic bound in Assumptions: $$X_0 \;\mapsto\; \{\;\mathsf{val} : \mathsf{Nat},\; \mathsf{inc} : \mathsf{Nat} \to X_0\;\}$$ Under 1-step legacySameType, comparing $X_0$ against its twice-unfolded expansion $\{\;\mathsf{val} : \mathsf{Nat},\; \mathsf{inc} : \mathsf{Nat} \to \{\;\mathsf{val} : \mathsf{Nat},\; \mathsf{inc} : \mathsf{Nat} \to X_0\;\}\;\}$ failed! This forced lean/Obend/Core/BidirectionalCompleteness.lean to carry a negative theorem full_corollary_fails showing that a legitimate declarative PartialTyping derivation could be rejected by checkAgainst when recursive bounds were unfolded to different depths.

The 8-Module Amadio–Cardelli Finite-Graph Bisimulation Spine

Ember replaced sameType with an executable, sound, and complete Amadio–Cardelli equi-recursive decision procedure across eight dedicated modules under lean/Obend/Core/:

ModulePathKey Definitions & Theorems (0 sorryAx)
FiniteBisimulationlean/Obend/Core/FiniteBisimulation.leanGeneric cycle-detecting reachability engine over finite state spaces (Finset S), proving visitedSame_iff and Equivalence (reflexivity, symmetry, transitivity of greatest bisimulation).
RecursiveRowslean/Obend/Core/RecursiveRows.leanDefines semantic row observations: visible head field lookup (lookup), tail kind (emptyRow vs rigid variable), and ordered resourceful shadow streams (shadowStream) so that shadowed non-shareable fields remain bisimulation-observable at every unfolding depth!
TypeGraphlean/Obend/Core/TypeGraph.leanBuilds the finite subterm closure of (actual, expected, assumptions.bounds) and proves candidate_iff: candidate assumptions a b = true ↔ Equivalent assumptions a b.
TypeGraphAliaslean/Obend/Core/TypeGraphAlias.leanProves alias_equivalent: every declared bound (i, bound) ∈ assumptions.bounds satisfies Equivalent assumptions (.variable i) bound, even when i occurs guarded inside bound.
FieldCongruencelean/Obend/Core/FieldCongruence.leanProves field_congruence: adding or replacing a field preserves Equivalent across arbitrary cyclic tails.
FiniteRowLawslean/Obend/Core/FiniteRowLaws.leanProves finite_field_commute ($f_1 \neq f_2 \Rightarrow \{f_1:A, f_2:B \mid r\} \sim \{f_2:B, f_1:A \mid r\}$) and finite_field_shadow (when the shadowed field is shareable).
CanonicalExtensionlean/Obend/Core/CanonicalExtension.leanProves legacySameType_extends: every pair accepted by the legacy canonicalizer is in Equivalent.
EquiRecursiveAgreementlean/Obend/Core/EquiRecursiveAgreement.leanExports agrees, agrees_iff, agrees_eq_candidate, agrees_equivalence, and all structural congruence lemmas, wired into Obend.Core.Check.sameType.

With sameType := EquiRecursiveAgreement.agrees wired into Obend.Core.Check, lean/Obend/Core/BidirectionalCompleteness.lean flipped full_corollary_fails into the positive completeness theorem full_annotated: every declaratively well-typed PartialTyping term with representative annotations is accepted by checkAgainst (0 sorryAx), while preserving progress_and_preservation and checked_execution_soundness in lean/Obend/Core/Soundness.lean.

2.4 The Algebraic Subtyping Investigation (review/ALGEBRAIC-SUBTYPING-20261031.md)

Alongside equi-recursive equality, Ember investigated whether Objective Bend should adopt Dolan's (2017) Algebraic Subtyping (MLsub) or Parreaux's (2020) Simple-sub (biunification over polar type lattices with $\top, \bot, \sqcup, \sqcap, \mu$). The complete mathematical analysis is preserved in /Users/ember/dev/obend-astra-decisions/review/ALGEBRAIC-SUBTYPING-20261031.md:

  1. Where Biunification Helps: Every Core4 type constructor except .specification and .prototype has clean polarity:
  2. arrow r q A B: contravariant in $A$, covariant in $B$, covariant in Reuse ($\mathsf{reusable} \le \mathsf{once}$), contravariant in Quantity along the usage-allowance order ($\mathsf{erased} \sqsubseteq \mathsf{linear}, \mathsf{affine} \sqsubseteq \mathsf{unrestricted}$).
  3. field l A R and variant R: covariant in member $A$ and tail $R$, with dual width subtyping (records drop fields going up: $\{l:A \mid R\} \le_{\text{rec}} R$; variants inject cases going up: $R \le_{\text{var}} [l:A \mid R]$).
  4. computation P R A: covariant in emitted plan $P$, contravariant in environment response $R$, covariant in result $A$: $$\frac{P \le P' \qquad R' \le R \qquad A \le A'}{\mathsf{computation}\;P\;R\;A \;\le\; \mathsf{computation}\;P'\;R'\;A'}$$
  5. Why mix and fix Break Plain MLsub Without Guarded Recursive Rows: As shown in ALGEBRAIC-SUBTYPING-20261031.md, in specification M (Self → Super → Row), the extension type has Self and Super in contravariant domain positions, while fix ties Self to the covariant output Row (Row ≤ Self). Under biunification, fix induces a polar cycle $\alpha^+ \le \{\text{fields}\}^+ \le \alpha^-$ that simplifies directly to an equi-recursive $\mu$-type $\mu\alpha.\{\text{fields}(\alpha)\}$—which is why the equi-recursive graph engine (Obend.Core.TypeGraph) is the exact prerequisite for polar subtyping.
  6. The Three Hard Boundaries Against Naive Width Subtyping:
  7. Resourceful Width Subtyping (linear / custody): Dropping a field via record width subtyping $\{a:\mathsf{Custody}(0), b:\mathsf{Nat}\} \le \{b:\mathsf{Nat}\}$ leaks the linear custody token! Width forgetting is sound only on fields where member.shareable = true—which is the exact invariant enforced by Obend.Elab.Coerce (lean/Obend/Elab/Coerce.lean).
  8. Reflective Transparency (prototype S T): Because reflect : prototype S T → S returns $S$ covariantly while prototype : S → T → prototype S T consumes $S$, prototype S T must remain invariant in $S$ whenever reflection is permitted, or else be split into a write/reflect view vs. a read-only projected view.
  9. Shadowed Field Subtyping under Late extend: If an object $\{a:\mathsf{Nat}, b:\text{self}.a\}$ is coerced by dropping $a$ before fix closes self, late binding of self.a dangles. Thus subtyping coercions apply to closed values (prototype/record) after fix, whereas open specifications retain rigid row tails.

3. Architecture of the 11 Interlocking Lean 4 Spine Modules

At the heart of /Users/ember/dev/obend/lean/Obend/ are 11 top-level spine modules that connect .obend source code, open-recursive templates, contextual truth-maintenance slots, cryptographic receipts, six-occasion laws, nominal kinds, pushout documents, and the guarded multi-runner ledger.

flowchart LR
    subgraph FrontEnd["Surface & Elaboration"]
        SPEC["Obend.Spec<br/>(.obend Parser, C4, D1–D6 Elaborate)"]
        ELAB["Obend.Elab<br/>(Prepare, Coerce η-Wrappers, Gate)"]
    end

    subgraph Core["Core4 Kernel"]
        CORE["Obend.Core<br/>(Term, Ty, EquiRec Check, Bidirectional, Machine)"]
    end

    subgraph Truth["Contextual Truth Maintenance"]
        SLOT["Obend.Slot<br/>(Supports, Judge, Reading, World, Selection)"]
        INT["Obend.Intention<br/>(Named Refs: Pinned vs Following)"]
    end

    subgraph Audit["Audited State & Proofs"]
        JUDGE["Obend.Judge<br/>(Graded Certs, Format, Verified LRAT)"]
        FOLD["Obend.Fold<br/>(Single-Pass Admission & Rely-Guarantee)"]
        REC["Obend.Receipt<br/>(cSHAKE256 Merkle, Shape, Trust, Dial)"]
    end

    subgraph Domain["Objects, Laws, Docs & Execution"]
        LAW["Obend.Law<br/>(6 Occasions, C4 Conjunction, Machine)"]
        KIND["Obend.Kind<br/>(Lineage Pins, create --from, Nominal Tooth)"]
        DOC["Obend.Doc<br/>(Patch Groupoid, Pushout Merge, Sections)"]
        RUN["Obend.Run<br/>(Guarded Runner, Control, Shared, Durable, Fork)"]
    end

    SPEC --> ELAB --> CORE
    SLOT --> INT
    INT --> SPEC
    INT --> DOC
    INT --> KIND
    CORE --> LAW
    CORE --> RUN
    JUDGE --> SLOT
    JUDGE --> LAW
    FOLD --> REC
    FOLD --> RUN
    LAW --> KIND
    SLOT --> RUN

3.1 Obend.Slot (lean/Obend/Slot.lean) — Contextual Truth-Maintenance Holes

Obend.Slot bridges Obend.Slot.Protocol (free indexed monads Prot S Hole A i with static holes) to a live changing World (Obend.Slot.World, lean/Obend/Slot/World.lean). Every deferred decision is a Slot Fact Cand (lean/Obend/Slot.lean):

lean
-- /Users/ember/dev/obend/lean/Obend/Slot.lean:37-81
structure Slot (Fact Cand : Type) where
  supports : Cand → List Fact
  judge : Cand → Bool

inductive Reading (Fact Cand : Type)
  | empty
  | rejected (c : Cand)
  | usable (c : Cand) (evidence : List (Fact × Bool))
  | blocked (c : Cand) (dead : List Fact)

@[expose] def Slot.read (S : Slot Fact Cand) (holds : Fact → Bool) : Option Cand → Reading Fact Cand
  | none => .empty
  | some c =>
    if S.judge c then
      match S.dead holds c with
      | [] => .usable c (S.evidence holds c)
      | d => .blocked c d
    else .rejected c

Four fundamental theorems (each guarded against vacuity by #tooth ... by mutation, lean/Obend/Slot.lean) govern every slot read:

  1. read_usable_supports (Slot.lean:130-132): (∃ ev, S.read holds (some c) = .usable c ev) ↔ S.judge c = true ∧ ∀ f ∈ S.supports c, holds f = true.
  2. read_blocked_exact (Slot.lean:135-138): S.read holds (some c) = .blocked c' d → d ≠ [] ∧ ∀ f, f ∈ d ↔ f ∈ S.supports c ∧ holds f = false (a blocked read names every dead support and nothing else, at least one).
  3. permanent_blocked_forever (Slot.lean:142-155): If a support $f \in S.\text{supports}(c)$ is Permanent holdsAt later f (such as a retired cell or an answered one-shot ring), then once $f$ goes false at view $v$, $c$ reads .blocked c d with $f \in d$ at every later view $v'$.
  4. read_of_usable (Slot.lean:158-165): Selecting a usable candidate $c'$ immediately reads .usable c' (S.evidence holds c') without reinstalling the slot's supports or judge.

In lean/Obend/Slot/World.lean, a World V Fact equips cells with versioned, content-addressed pins (PinId = ByteArray, computed via cSHAKE256 in Obend.Receipt.CShake256) and proves:

  • retired_dead: Retiring a cell permanently falsifies .live cell.
  • ver_step: Revising a cell increments its version strictly.
  • stale_behind_forever: A support citing a superseded pin remains false at every later view unless a hash collision (Collides) occurs.

In lean/Obend/Slot/Selection.lean and lean/Obend/Slot/Integrity.lean, SelectionSound and incumbentTransition prove that incumbent-preserving slot selection never switches away from a still-usable incumbent (usable_incumbent_kept) and switches to a fallback only when the incumbent is blocked (switched_only_when_incumbent_blocked).

3.2 Obend.Intention (lean/Obend/Intention.lean) — LTUO §9.3.7 Named References as Slots

In LTUO §9.3.7, Faré defines an Intention as a named reference that records how the author intends it to track upstream evolution (e.g., pinned to a specific content hash vs. following a branch/cell head). Obend.Intention proves that an Intention is not a separate primitive: it is a Slot Fact Pin whose candidates are pins of the target cell (lean/Obend/Intention.lean):

lean
-- /Users/ember/dev/obend/lean/Obend/Intention.lean:36-93
structure Intention (W Ref Pin Fact Err : Type) where
  resolve : W → Ref → Except Err Pin
  supports : Ref → Pin → List Fact
  judge : Ref → Pin → Bool
  holds : W → Fact → Bool

def Intention.slot (I : Intention W Ref Pin Fact Err) (r : Ref) : Slot Fact Pin where
  supports := I.supports r
  judge := I.judge r

def Intention.stale (I : Intention W Ref Pin Fact Err) (w : W) (r : Ref) (p : Pin) : List Fact :=
  (I.slot r).dead (I.holds w) p

Its four core theorems (Intention.lean:111-189) establish:

  • stale_iff_blocked: When I.judge r p = true, I.read w r (some p) = .blocked p d ↔ d ≠ [] ∧ d = I.stale w r p.
  • resolve_fill: If I.resolve w r = .ok p and ValidResolve holds, then (I.slot r).read (I.holds w) (some p) = .usable p ((I.slot r).evidence (I.holds w) p).
  • renew_repairs: I.renew w r = .ok p (re-running I.resolve w r when a following reference falls behind) produces a pin that reads .usable immediately.
  • stays: For any view transition w → w' that preserves the supports of (r, p), a usable pin p remains .usable at w'.

Obend.Intention is instantiated in three distinct places across the codebase:

  1. Package extends references (Obend.Spec.Ecosystem, lean/Obend/Spec/Ecosystem.lean).
  2. Document section citations (Obend.Doc.Section, lean/Obend/Doc/Section.lean).
  3. Nominal kind parent lineages (Obend.Kind.kindRefs, lean/Obend/Kind.lean).

3.3 Obend.Spec (lean/Obend/Spec.lean) — The .obend Front End, C4 Linearizer, and Modular Template Elaborator

Obend.Spec compiles .obend source files into checked Obend.Core.Term programs plus a typing proposal verified by Obend.Core.Bidirectional.checkAgainst.

Concrete .obend Surface Syntax (stdlib/std.obend and Spec/Syntax.lean)

A .obend program consists of optional package headers (package NAME [extends REF ...]), law blocks, let bindings, template declarations (template (self : Self, super : Super) [metadata M] [law L] { field = expr, ... }), mix, fix, prototype, reflect, project, extend, split, perform, and done. For example, /Users/ember/dev/obend/stdlib/std.obend defines reusable standard-library specifications:

obend
// /Users/ember/dev/obend/stdlib/std.obend (excerpt)
let counter_base =
  template (self : Self, super : Super)
  metadata { name = "std.counter.base", version = 1 }
  {
    count = 0,
    step  = 1,
    read  = self.count,
    bump  = self.count + self.step
  }
in
let counter_double =
  template (self : Self, super : Super)
  metadata { name = "std.counter.double", version = 1 }
  {
    step = super.step * 2
  }
in
let spec = mix counter_base counter_double in
let proto = prototype spec (fix spec { count = 0, step = 1, read = 0, bump = 0 }) in
proto.bump

Modular Template Typing (Rules D1–D6) and Composition Contracts (Spec/Elaborate.lean, Spec/Contract.lean)

As documented in paper/objective-bend.md and implemented in lean/Obend/Spec/Elaborate.lean and lean/Obend/Spec/Contract.lean:

  • Every template (self, super) is elaborated once, at definition site, against two distinct rigid row variables $X_{\text{self}}$ and $X_{\text{super}}$ whose lower bounds in Assumptions.bounds record only the fields read from self.f and super.g inside that template's body (Rules D1–D6).
  • Obend.Spec.Contract proves that layer contracts form a monoid under mix (run_append, run_assoc, checkLayer_ok, discharge_ok), and Obend.Spec.Proofs.Compose proves composeTy_closed and compose_typed: Specification<T> is closed under compose without re-inspecting template bodies at fix time!
  • End-to-end front-end soundness is proved in Obend.Spec.Proofs:
  • accepted_typed: Every .obend source accepted by FrontEnd.accept has a valid PartialTyping derivation in Obend.Core.Check.
  • accepted_never_refused: The bounded demand machine never halts in MachineError on an accepted program.
  • accepted_execution_semantics: A finished demand-machine run extracts the exact deep source evaluation (DeepEval) of the elaborated term.
  • decode_json / decodePacket_term: The wire decoder inverts the front end's canonical wire encoding, so a receiving node checks the exact Term rather than re-parsing untrusted source text.

LawDials and Bellman-Ford Law Repair Synthesis (Spec/LawDial.lean)

When composing multiple inheritance layers whose per-field laws impose numeric difference constraints $x_i - x_j \le c_{ij}$ (for example, min_balance <= balance in one layer and balance <= cap in another), lean/Obend/Spec/LawDial.lean provides:

  • confluent_order_independent: Conjunction of pure predicate laws over C4 is commutative, associative, and idempotent up to boolean equivalence.
  • synthesizeLawRepair: Runs a verified Bellman-Ford shortest-path / negative-cycle detector over difference-constraint laws:
  • synthesizeLawRepair_sound: Every synthesized repair assignment satisfies all conjoined layer constraints.
  • synthesizeLawRepair_complete_refutation: When synthesizeLawRepair refuses, it emits a negative-weight cycle witness proving that the conjoined laws are unsatisfiable over $\mathbb{Z}$.

3.4 Obend.Elab (lean/Obend/Elab.lean) — Annotation Preparation, Coercions, and the 16-Cell Quantity Matrix

Obend.Elab sits between unannotated/partially-annotated terms and Obend.Core.Bidirectional:

  1. Obend.Elab.Prepare (prepare_checks): lean/Obend/Elab/Prepare.lean implements prepare : Assumptions → Context → Term → Annotations → Ty → Annotations, which walks any declaratively well-typed Checkable term and fills in every representative annotation required by Bidirectional.checkAt. Theorem prepare_checks proves that checkAgainst succeeds on (prepare ...) (0 sorryAx).
  2. Obend.Elab.Coerce (coerce_checks and the 16-Cell Quantity Matrix): lean/Obend/Elab/Coerce.lean synthesizes explicit Core4 $\eta$-expansion wrappers witnessing subtyping coercions Sub assumptions A B without changing the kernel checker!
  3. Dropping a shareable record field $f$ elaborates to a record reconstruction reading the kept fields.
  4. Coercing a function type (r₁, q₁, A₁ → B₁) ≤ (r₂, q₂, A₂ → B₂) wraps $f$ in $\lambda x :_{q_2} A_2.\; \text{coerce}_{B_1 \le B_2}(f\;(\text{coerce}_{A_2 \le A_1}\;x))$.
  5. Because the wrapper binds $x$ at $q_2$ and passes it to $f$ expecting $q_1$, Coerce.lean proves the exact 16-cell quantity compatibility matrix (qle : Quantity → Quantity → Bool, verified completeness by counterexample on all 9 false cells in lean/docs/REPORT-ELAB-20261031.md):
q₁ \ q₂ (Domain Expects \ Wrapper Binds)erasedaffinelinearunrestricted
erasedfalsetruefalsetrue
affinefalsetruefalsetrue
linearfalsetruetruetrue
unrestrictedfalsefalsefalsetrue

(Why is (erased, erased) false for an $\eta$-wrapper $\lambda x.\; f\;x$? Because the wrapper's syntactic body f x mentions x once in the argument position, so an erased binder on the $\eta$-lambda fails binderOverused with count = 1 > 0!)

  1. Obend.Elab.Work (checkWork): lean/Obend/Elab/Work.lean defines the fuel-free structural recursion measure checkWork : Term → Nat (sizeOf), proving checkWork_pos and strict decrease on every recursive call of synthAt / checkAt.

3.5 Obend.Judge (lean/Obend/Judge.lean) — Graded Proof Checkers and Verified Foreign LRAT SAT Checking

Obend.Judge makes proof and certificate checking pluggable while preventing trivial (fun _ _ => true) judges from being registered:

lean
-- /Users/ember/dev/obend/lean/Obend/Judge.lean:38-80
class Judge (Obj Cert : Type) where
  Valid : Obj → Prop
  Invalid : Obj → Prop := fun o => ¬ Valid o
  Fragment : Nat → Obj → Prop := fun _ _ => True
  disjoint : ∀ {o}, Valid o → Invalid o → False := by intro _ hv hi; exact hi hv
  check : Nat → Obj → Cert → Decision
  sound : ∀ {k o c}, check k o c = .proved → Valid o
  refuted : ∀ {k o c}, check k o c = .refuted → Invalid o
  complete : ∀ {k o}, Fragment k o → Valid o → ∃ c, check k o c = .proved
  mono : ∀ {k k' o c d}, k ≤ k' → d ≠ .inconclusive → check k o c = d → check k' o c = d
  • trivial_unregistrable (Judge.lean:216-221): Every registered Format Obj must carry a planted invalid witness (witnessObj, witnessCert) that its decoder + checker does not accept (witness_not_proved), making fun _ _ _ => .proved unregistrable at the type level.
  • Verified Foreign LRAT Checker (Judge/Lrat.lean, Judge/Identity.lean, Judge/Certified.lean): In lean/Obend/Judge/Lrat.lean, Ember implemented a total, verified LRAT (Linear Reverse Asymmetric Tautology) unsatisfiability certificate checker with lrat_sound proved against boolean assignment semantics (0 sorryAx). In lean/Obend/Judge/Certified.lean, an external CaDiCaL 3.0.1 SAT solver emitted a 215-step LRAT proof (noCarryLrat, 272 clauses over 97 variables) for the 8-bit noCarry arithmetic identity (a + b == a ^ b) ↔ (a &&& b == 0); Lean's kernel verifies the foreign LRAT certificate via decide +kernel, discharging noCarry_holds : ∀ a b : BitVec 8, ...!

3.6 Obend.Fold (lean/Obend/Fold.lean) & Obend.Receipt (lean/Obend/Receipt.lean) — Single-Pass Audited Folds and cSHAKE256 Epistemic Receipts

In Objective Bend, batch verification of a state log, a document history, or a multi-runner ledger is a single left fold List.foldl parameterized by an Admission W Rest Row Fact Clause (lean/Obend/Fold.lean):

  • receipt_sufficient (Fold.lean:160-169): Any two worlds w and w' that agree on the pins read by row r (∀ c ∈ A.reads r, w.pinOf c = w'.pinOf c) step identically (A.step (w, rest) r = .ok (w, rest') ↔ A.step (w', rest) r = .ok (w', rest')). Thus a verifier never needs the full World—only the Merkle openings of A.reads r!
  • project_moves_run (Fold.lean:245-261): Proves Jones-style rely–guarantee decomposition over a shared Family: if other principals' moves satisfy OtherStep (preserving this principal's relied-upon invariants), projecting a joint multi-principal fold onto principal $i$ yields a valid local Run of $i$.

Each admitted row issues a Receipt Cell Pin Fact Clause (lean/Obend/Receipt.lean) backed by a positional FIPS 202 cSHAKE256 Merkle tree (Obend.Receipt.CShake256 and Obend.Receipt.Merkle, lean/Obend/Receipt/Merkle.lean, proved collision-binding in opens_binds). Every receipt carries three explicit epistemic dimensions:

  1. Shape (Receipt.lean:274-292): Records the quantifier alternation of the verified claim, ordered linearly from weakest to strongest: $$\mathsf{existsRun}\;(\exists)\;<\;\mathsf{planForAll}\;(\exists\text{plan}\,\forall\text{env})\;<\;\mathsf{policyForAll}\;(\exists\text{policy}\,\forall\text{env})\;<\;\mathsf{forAllPlan}\;(\forall\text{plan}\,\exists\text{run})\;<\;\mathsf{forAll}\;(\forall)$$ Theorem shape_strict (Receipt.lean:314-319) proves that each level strictly implies the previous and is separated by a concrete countermodel (exists_not_forAll).
  2. Trust (Receipt.lean:328-367): Tags each support as .checked (verified by a kernel Judge) or .attested (signed by an external oracle/principal). Theorem ceiling_checked proves that a receipt's overall trustCeiling is .checked if and only if every support it cites is .checked (∀ p ∈ r.checked, p.2 = .checked).
  3. Dial (Receipt.lean:376-403): Controls selective disclosure (full vs. cited vs. bit). Theorem dial_invariant proves that redacting a receipt via r.disclose d never alters its shape, trustCeiling, or outcome!

3.7 Obend.Law (lean/Obend/Law.lean) & Obend.Kind (lean/Obend/Kind.lean) — Six-Occasion Laws, C4 Conjunction, and First-Class Nominal Kinds

To close fareoo's K-LAW-INHERIT and K-KIND-CELL gaps, Obend.Law and Obend.Kind formalize laws and nominal kinds directly over Core4:

  • Six Occasions (Occasion, Law.lean:45-64): Every cell transition fires on one of six occasions:
lean
  inductive Occasion where
    | birth | adopt | relaw | retire | write | invoke
    deriving DecidableEq, Repr
  • Laws as Core4 Terms (lawTy S, Law.lean:77-78): A law over state row S is an ordinary Core4 term of type Ctx(S) → Bool, where Ctx(S) is the 7-field record { occasion : Label, who : Label, before : S, after : S, action : Label, spend : Nat, epoch : Nat }.
  • Standard Law Generators (Obend.Law.Std, lean/Obend/Law/Std.lean): Provides six composable combinators (ownerSigns, allowlist, bounded, machine, quorum, windowed) plus keeps : Law → Claim → Prop, proving guardedBy_sound and chain_keeps (an inductive invariant preserved across arbitrary admitted transition chains).
  • C4 Conjunction Inheritance (K-LAW-INHERIT Closed, Law.lean:268-401): Given C4 ancestry anc : List (String × Term) from root to self, effectiveLaw anc compiles to a right-associated Core4 .ifBool short-circuit conjunction. Three theorems seal K-LAW-INHERIT:
  • effective_law_eval: Evaluating effectiveLaw on the Core4 demand machine equals (anc.all fun (_, l) => evalLaw B l c = .ok true).
  • inherited_never_widens: Extending an ancestry anc with additional layers more can never turn a false verdict into true (evalEffective B anc c = .ok false → evalEffective B (anc ++ more) c = .ok false). A subkind or wrapper can only strengthen its ancestors' laws, never bypass them!
  • effective_refuses_first: When a transition is refused, checkEffective blames the first ancestor (by, l) in C4 root-to-self order whose law returned false, returning Refusal := { by, reason }.
  • First-Class Nominal Kinds and Lineage Staleness (Obend.Kind, Kind.lean:39-340): A KindCell lives in KindWorld with its own content-addressed KindPin (computeKindPin over name, parents, lawTag, shape, and full resolved lineage).
  • Why does KindPin hash the full lineage : List (String × ByteArray) rather than just direct parents? During the October 30 audit (review/KIMI-REVIEW-20261030.md), Kimi pointed out that if C extends B and B extends A, revising grandparent A without modifying B's declaration would leave B's pin unchanged if B only hashed parent names! Including lineage in KindPin enabled Ember to prove ancestor_revised_stale (Kind.lean:291-340): revising any proper ancestor $A$ in a kind's C4 DAG changes resolveLineage, changes computeKindPin, and immediately marks every downstream following kind reference stale!
  • Furthermore, nominal_tooth (Kind.lean:231-244) proves that two objects c₁ and c₂ with identical structural shape and body (c₁.shape = c₂.shape ∧ c₁.body = c₂.body) but distinct nominal kind tags ("Meters" vs "Feet") dispatch to different methods under Kind.dispatch (dispatch table c₁ = some "add-meters" vs some "add-feet").

3.8 Obend.Doc (lean/Obend/Doc.lean) — Patch Groupoid, Pushout Merge, and Documents That Remember Why

Obend.Doc turns structured documents into consensus objects with categorical patch history:

  • Typed Graph Patches (Obend.Doc.Patch, lean/Obend/Doc/Patch.lean): Document edits are typed operations Add, Delete, Connect, SetField over node/edge graphs, forming a partial action with composition (apply_compose), inversion (apply_invert: invert q undoes q on any document where q succeeds), and framing (apply_frame: patches on disjoint footprints commute).
  • Pushout Merge and Conflict Characterization (Obend.Doc.Merge, lean/Obend/Doc/Merge.lean): Over the thin category Le of non-conflicting patch extensions (Obend.Doc.Category), Doc.merge is proved to be the categorical pushout:
  • merge_isPushout & pushout_unique: When two branches $p, q$ do not conflict (conflicts p q = []), merge p q is the unique least upper bound in Le.
  • no_cocone: When conflicts p q ≠ [], no cocone exists in Le (∀ u, ¬(Le p u ∧ Le q u)).
  • files_no_pushout: Formalizes Mimram & Di Giusto's (2013) diagram (3) showing that line-based file patches do not have pushouts (two compatible upper bounds with no common refinement), justifying why Objective Bend versions typed node/edge/field graphs rather than raw text lines!
  • Sections as Slots & Backlinks (Obend.Doc.Section, Obend.Doc.Backlink, Doc.lean:155-268): Every document section is a Slot whose candidates carry prose, code, and citations (Intentions to other sections or proof objects). When merge encounters a conflict on key k, conflict_is_section turns the conflict into a live Section holding both competing candidates [v₁, v₂] rather than aborting the document (conflict_section_usable). And usable_unmoved proves that editing any node outside a section's cited support cone leaves the section's .usable reading untouched.

3.9 Obend.Run (lean/Obend/Run.lean) — The Guarded Runner, Selection Control, Shared Ledger, Crash Recovery, and Offline Fork/Stitch

Obend.Run executes open objects whose methods yield Activity<Plan, Response, Result> computations (Term.perform plan):

  1. Guarded Dispatch & Replanning (Obend.Run.Runner, Obend.Run.Proofs.Invariant):
  2. refines: Every reachable configuration of Runner.run satisfies Inv, so its accepted event log is Legal (legal_iff_admission).
  3. inflight_replan_refused & lost_reply_frozen: While an external call is in flight, replanning that slot is refused (busy); if a reply is lost (abandon), the obligation remains recorded in the log (abandon_keeps_obligation).
  4. Selection Control (Obend.Run.Control, lean/Obend/Run/Control.lean):
  5. Every hole resolution checks the caller's principal against the slot's chooser office and applies one of three verified presets (firstUsable, incumbentFirst, chooserRequired).
  6. Tenure (whileInOffice vs lasting): Under .whileInOffice, revoking a chooser's office immediately invalidates selections resting on .holdsOffice who (office_revoked_blocks_whileInOffice), whereas .lasting selections survive office rotation provided their domain supports still hold (office_revoked_keeps_lasting).
  7. Multi-Runner Shared Ledger (Obend.Run.Shared, lean/Obend/Run/Shared.lean):
  8. Arbitrates any number of controlled runners over a shared World through an atomic commit boundary.
  9. joint_reachable_ticket_exclusive & loser_refused_names_winner_and_epoch: A one-shot capability token (OnceSpent) can be spent by at most one runner across the entire joint execution, and the losing runner receives a structured refusal naming the winning principal and commit epoch.
  10. Journal-First Durable Crash Recovery (Obend.Run.Durable, lean/Obend/Run/Durable.lean):
  11. Formalizes a host with volatile memory and a durable append-only journal + outbox where the journal append is the linearization point.
  12. recover_restores_memory & exposed_once: Across arbitrary crash-and-replay sequences, volatile state is restored to the last journaled commit and any external effect consuming a single token is exposed to the outbox at most once.
  13. Offline World Fork & Atomic Stitch (Obend.Run.Fork, lean/Obend/Run/Fork.lean):
  14. An offline worker forks the World (Fork.fork); monotone rows merge by union, while bounded/exclusive rows escalate unjudged.
  15. When reconnecting, Fork.stitch replays escalated dispatches through the joint runner+ledger transition atomically: stitch_is_commit proves accepted stitches equal online commits, stitch_double_spend proves a token spent both on main and on the offline fork is refused at stitch time naming the online winner and epoch, and every refused offline branch is preserved as a conflict Doc.Section (stitch_refusal_kept_as_section).

4. The Late-October / November 1, 2026 Architectural Debates in obend-astra-decisions

The repository /Users/ember/dev/obend-astra-decisions preserves the complete decision archive, formal lane briefs (59 LANE-*-BRIEF-*.md and REPORT-*.md files in lean/docs/), adversarial model reviews (review/), and Astra synthesis documents (astra/) from the intense October 29 – November 1, 2026 verification sprint. Below we reconstruct the seven defining technical debates and their formal resolutions.

4.1 Astra Effect Laws (astra/ASTRA-EFFECT-LAWS-20261031.md) and Guarded Law Judgments (Obend.Run.LawJudge)

In /Users/ember/dev/obend-astra-decisions/astra/ASTRA-EFFECT-LAWS-20261031.md, Ember and Astra confronted how object laws interact with effectful activities (Term.perform plan in Obend.Run):

  • The Problem: In minidregg, a cell law was only a state predicate Ctx(S) → Bool evaluated synchronously on state writes. When Objective Bend introduced Activity<Plan, Response, Result> (Ty.computation), an object could yield an external Plan (such as transferring custody or invoking a remote capability) across multiple asynchronous turns. If laws only checked completed state writes, an illegal external Plan could escape to the outbox before its state write occurred!
  • The Resolution (Obend.Run.LawJudge and Obend.Law.Machine):
  • Every emitted Plan is judged at dispatch time (Occasion.invoke) before leaving the runner (Obend.Run.LawJudge, lean/Obend/Run/LawJudge.lean), passing through the full C4 effectiveLaw conjunction of the acting cell and its kind (dispatch_checked_by_effective_law).
  • When a reply returns (or is abandoned), the continuation transition is re-judged against the updated world epoch (resume_rechecked_at_epoch), preventing time-of-check to time-of-use (TOCTOU) drift across asynchronous yields.
  • In lean/Obend/Law/Machine.lean (REPORT-LAWFINISH-20261031.md), stateful multi-step protocol laws are compiled into deterministic finite-state probes (Machine.probe) with proved judge_sound and judge_complete instances for Obend.Judge.

4.2 The Contextual Hole Repair Exercise (astra/HOLE-REPAIR-SET-20261030.md)

On October 30, 2026 (commit d8db9f8), Ember tested a major metatheoretic hypothesis: Can contextual modal holes (Term.hole n σ, with delayed explicit substitutions $\sigma : \text{List Term}$) be added directly to the 25-constructor Core4 AST without shattering the existing 15,000-line soundness and bidirectional completeness proofs?

As documented in /Users/ember/dev/obend-astra-decisions/astra/HOLE-REPAIR-SET-20261030.md:

  • First, Ember factored a generic binder-aware functorial traversal Obend.Core.Traverse (lean/Obend/Core/Traverse.lean), proving that Term.rename and Term.substitute are instances of one structural traversal (traverse).
  • When Term.hole (n : Nat) (σ : List Term) was added as constructor #1 of Term, 572 kernel declarations had their types or terms revised, yet only 20 authored proofs across 10 files required manual human/agent repair! Every other lemma across Core/, Spec/, and Law/ discharged automatically via the uniform Traverse induction principle.
  • In lean/Obend/Core/Hole.lean, Ember proved the CMTT contextual substitution commutation laws (rename_fillHole, substitute_fillHole, and typing_fillHole), establishing that Obend.Slot's runtime hole resolution and Obend.Core.Term's syntactic .hole n σ share a single formal foundation.

4.3 The Justification Semiring Refutation & The Parsimony Deletion (REPORT-JUST-20261031.md, REPORT-PARSIMONY-20261101.md)

One of the most striking episodes in obend-astra-decisions is the rise, formal refutation, and surgical deletion of the justification polynomial semiring (Obend.Just and Obend.Slot.PolySlot).

  1. The Original Hypothesis (LANE-JUST-BRIEF-20261031.md): Early design notes proposed generalizing Slot's flat support list supports : Cand → List Fact to positive boolean provenance polynomials $\mathbb{N}[X]$ quotiented by Distributive Lattice (DL) equations ($a + a = a$, $a \cdot a = a$, $a + a \cdot b = a$), claiming that (a) absorption $a + a \cdot b = a$ implies multiplicative idempotence $a \cdot a = a$, (b) automated truth-maintenance repair always finds a single-pin replacement (T2), and (c) factoring Slot, Intention, and Receipt through Just.lean would shrink downstream proof obligations.
  2. Astra's Two Formal Countermodels (lean/docs/REPORT-JUST-20261031.md):
  3. Algebraic Countermodel (absorption_not_mul_idempotent): Astra constructed a 3-element commutative semiring $\{0, 1, 2\}$ where addition is idempotent and absorption $a + a \cdot b = a$ holds everywhere, yet $2 \cdot 2 = 0 \neq 2$! Thus mul_idem had to be added as an independent axiom of Why polynomials.
  4. Repair Soundness Countermodel (clash_unsat): Astra proved that naive repair completeness (T2) is false when a candidate's justification polynomial conjoins two distinct pins of the same cell ($c@p_1 \wedge c@p_2$ with $p_1 \neq p_2$), because a single cell in World has a unique current pin w.pinOf c! Thus repair soundness holds if and only if the support conjunction satisfies the pairwise pin-compatibility invariant Repair.Compatible.
  5. The Parsimony Audit and Deletion (commit f1e1406, lean/docs/REPORT-PARSIMONY-20261101.md): On November 1, 2026, Ember wrote tools/proofsize.py to measure whether factoring Slot through Just.lean actually shortened the proof spine. The script revealed that the polynomial-factored spine was 5 lines longer (and added 1,872 lines of semiring canonicalizer machinery) than the direct List Fact formulation! Following Occam's razor, commit f1e1406 (LANE-PARSIMONY) deleted Just.lean, PolySlot.lean, and Slot/Repair.lean (-1,872 net lines), preserving only the battle-tested List Fact spine in Obend.Slot.

4.4 Clairvoyant Call-by-Value vs. Lazy Self-Referential Records (REPORT-COST-20261101.md)

In LTUO and DESIGN.md, a recurring question was whether Core4's lazy call-by-name mix/fix evaluation could be cost-modeled by Hackett & Hutton's (2019) clairvoyant call-by-value semantics (which models call-by-need sharing by non-deterministically choosing at each binder whether to evaluate the argument once or skip it if unused).

In lean/Obend/Core/Cost.lean and lean/Obend/Core/Clairvoyant.lean (reported in lean/docs/REPORT-COST-20261101.md):

  1. Positive Theorem (mix_assoc_cost): On the Core4 demand machine (Obend.Core.Machine), re-associating a 3-layer mixin composition mix (mix a b) c vs. mix a (mix b c) preserves the exact result and changes the machine tick count by at most an additive constant of 8 ticks (CostWithin 8, 2 ticks per mix frame push/pop).
  2. Negative Refutation (self_field_no_cv and ticks_not_within_cv): When Astra formalized Hackett–Hutton clairvoyant CBV over Core4's open-recursive records, it discovered that whole-record clairvoyant CBV has NO derivation at all for { a = 7, b = self.a }.b!
  3. Why? To evaluate fix spec base eagerly to a record value under whole-record CBV, self must be demanded before spec self base returns the record { a = 7, b = self.a }, creating a strict cycle on self even when b only reads the already-convergent sibling field self.a!
  4. By contrast, on Core4's lazy demand machine, { a = 7, b = self.a }.b converges in 25 ticks because fix substitutes fix spec base by name and .get ... "a" only forces field a, never forcing the entire record at once.
  5. Consequently, Clairvoyant.lean proves self_field_no_cv (no clairvoyant CBV cost exists for self-sibling record projection) and restricts the clairvoyant cost correspondence to the non-self-sibling fragment (cv_sound_non_self).

4.5 The Symbolic Mathematics Substrate (NEXT-MATHEMATICA.md and Obend.Cas / Obend.Math)

In /Users/ember/dev/obend/NEXT-MATHEMATICA.md (253 lines), Ember articulated why Objective Bend's combination of open-recursive specifications, Judge-verified certificates, and Doc sections is the natural substrate for a verified symbolic computation & computer algebra system (CAS) surpassing untrusted rewrite systems like Mathematica:

  • The Diagnosis of Traditional CAS: In Mathematica or SymPy, symbolic simplification rules are global, unverified pattern-rewrite tables; adding a domain assumption (e.g., $x > 0$ or $R$ is a commutative ring) mutates global context or forks the simplifier, and when a branch cut or pole invalidates an identity, notebook cells downstream silently display stale formulas.
  • Objective Bend's CAS Architecture (lean/Obend/Cas/ and lean/Obend/Math/):
  • Algebraic Theories as Open Specifications: Ring, field, differential, and polynomial normalizers are open templates composed via mix; domain assumptions are Slot supports (Fact).
  • Rewrites Judged by Certificates (Obend.Judge): Every non-trivial symbolic step (polynomial ideal membership, Positivstellensatz witness, truth-table / LRAT unsatisfiability, interval enclosure) emits a certificate checked by a registered Judge, recording its quantifier Shape and Trust ceiling in an Obend.Receipt.
  • Live Notebooks as Obend.Doc: Equations and derivations are Doc.Section slots citing upstream lemmas via Obend.Intention. If an upstream domain assumption is retracted or a parameter crosses a branch cut, every dependent symbolic derivation transitions immediately from .usable to .blocked naming the exact falsified assumption!

4.6 Consensus Contracts and DeOS / Dregg2 Grounding (DEOS-CONTRACTS-CONSENSUS.md and Obend.Dregg2 / Obend.DeOS)

In /Users/ember/dev/obend/DEOS-CONTRACTS-CONSENSUS.md (185 lines), Ember mapped how Objective Bend serves as the smart-contract and decentralized operating system (DeOS / Dregg2) kernel:

Traditional Smart-Contract / OS FailureObjective Bend Kernel MechanismProved Theorem & Module
Proxy Upgrade Hazards (delegatecall storage collision & law bypass)C4 law conjunction (effectiveLaw) + KindPin lineage hashing + carry state migrationinherited_never_widens (Obend/Law.lean:345), ancestor_revised_stale (Obend/Kind.lean:325)
Linear Asset Double-Spend / ReentrancyTy.custody + Quantity.linear + Term.split + Run.Shared ticket exclusivity + Run.Runner in-flight freezevalue_consumes_linear (Core/LinearSoundness.lean), joint_reachable_ticket_exclusive (Run/Proofs/Shared.lean), inflight_replan_refused (Run/Proofs/Invariant.lean)
Full-State Replay Requirement for Light VerifiersSingle-pass Fold.Admission + cSHAKE256 Receipt.Merkle read-pin openingsreceipt_sufficient (Obend/Fold.lean:160), recheck_sound (Obend/Receipt.lean:181)
Opaque Governance / Silent Key RotationRun.Control chooser offices with explicit Tenure (whileInOffice vs lasting)office_revoked_blocks_whileInOffice, office_revoked_keeps_lasting (Run/Proofs/Control.lean)
Offline Partition / Local-First Merge ConflictsRun.Fork offline execution + atomic Fork.stitch escalating conflicts into Doc.Sectionstitch_is_commit, stitch_double_spend (Run/Fork.lean)

4.7 The November 1, 2026 Multimodel Council Audits (ASTRA-COUNCIL-20261101.md and FABLE-COUNCIL-20261101.md)

On November 1, 2026, Ember convened two independent multimodel verification councils—recorded in /Users/ember/dev/obend/astra/ASTRA-COUNCIL-20261101.md and /Users/ember/dev/obend/review/FABLE-COUNCIL-20261101.md (synthesizing Kimi, Argon, and Fable)—to audit both the checkpoint-20261101 spine (commit 6266457, 300 Lean modules, 35,127 constants, 0 sorryAx) and the subsequent late-November-1 burst (commits ec3e54c..7c699ca, which expanded the tree to 452 Lean modules and 45,588 constants, 0 sorryAx).

The Councils' findings are remarkable for their precision:

  1. Zero Kernel Axioms & Non-Vacuity Tooth Guards: Across all 452 Lean modules (45,588 constants), sorryAx count is strictly zero (0), and the core spine uses #tooth <thm> by mutation <def> := ... (Obend.Tactic.Tooth, lean/Obend/Tactic/Tooth.lean) to mechanically verify that mutating the underlying definition breaks the theorem's proof, ruling out vacuous False → ... premises.
  2. Four Genuine Kernel Advancements in the Nov 1 Burst (ec3e54c..7c699ca):
  3. ef8f814 (REPORT-LINEAR-20261101.md): Added Term.split (multiplicative record destructor) and proved Obend.Core.LinearSoundness (valueStep_preserves_typing_and_linear1).
  4. a4959c0 (REPORT-KINDS-20261101.md): Bridged Obend.Kind into Core4's type checker by encoding a nominal kind tag K as a phantom #kind : Label(K) field in Ty.field, proving that structural width coercion (Obend.Elab.Coerce) refuses to coerce across distinct #kind tags when #kind is protected.
  5. c490226 (REPORT-QUOTE1-20261101.md): Oriented Obend.Quote.Eval so reflect (prototype s t) reifies s into a quotation record carrying its content hash and source AST (quote_reflect_sound).
  6. f0bb666 (Obend.Lower): Built a Tait–Girard reducibility predicate (Obend.Lower.Tait, lean/Obend/Lower/Tait.lean) proving strong normalization on the fix-free Core4 fragment and compiling closed first-order functions to affine BendTT interaction-net Book rules (Obend.Lower.Bridge).
  7. Critical Findings & Gaps Exposed by the Councils:
  8. Prose vs. Code Drift in DESIGN.md §1: FABLE-COUNCIL-20261101.md §2 pointed out that DESIGN.md still describes aspirational dependent type constructors (pi, sigma, id, refl, jElim, universe u) from an early BendTT design draft that do not exist in Obend.Core.Term or Obend.Core.Ty (which implement the simply-typed/row-polymorphic/equi-recursive Core4 calculus).
  9. Trust Boundary Probes in AstraDecisionProbe.lean: In /Users/ember/dev/obend-astra-decisions/AstraDecisionProbe.lean, Ember constructed machine-checked counterexample probes demonstrating two subtle trust boundaries:
  10. Export.verify Caller-Supplied Root (probe_export_accepts_any_dated): In Obend.Receipt.Export, Export.verify checks that a receipt's Merkle openings match the dated root passed to verify, meaning the caller must authenticate the dated state root out-of-band (or bind it to Run.Shared's ledger root).
  11. quoteVia Principal Agnosticism (probe_quoteVia_ignores_principal): In Obend.Quote, quoteVia checks that a capability grant exists in the table without binding the grant's Principal to the caller—a gap that was subsequently closed in Obend.Run.Reflect (lean/Obend/Run/Reflect.lean, theorem forged_descriptor_refused and metadata_grant_cannot_quote).
  12. Shallow Encoding in Late-Burst Domain Leaves: Both councils distinguished the deeply verified 300-module core spine (Core, Spec, Elab, Slot, Intention, Judge, Fold, Receipt, Law, Kind, Doc, Run, Leaf, Lower) from the late Nov 1 domain-application leaves (Compat, Vcs, Verify, Effects, Capabilities, Consensus, Silicon), noting that several late domain leaves model state transitions as finite struct/boolean predicates rather than embedding through Obend.Core.Term.

5. Rust Runtime Crates, Native FFI, Sandstorm Harness, and GUI Workbench

Beyond the 452 Lean 4 modules, /Users/ember/dev/obend ships a four-crate Rust workspace (Cargo.toml) and an interactive test harness that execute .obend programs and world vaults natively:

Crate / ComponentPathRole & Key Capabilities
obend-fficrates/obend-ffi/src/lib.rsLinks directly against the Lean 4 C runtime (libleanshared + compiled Obend static archive) to expose obend_eval_source, obend_check_packet, and obend_vault_step with zero re-implementation drift between Lean proofs and native execution.
obend-nodecrates/obend-node/src/main.rsContent-addressed daemon and CLI runner wrapping Obend.Vault (lean/Obend/Vault.lean): persists world snapshots, journaled commits, Merkle receipts, and .obend packages on disk.
obend-sandboxcrates/obend-sandbox/src/lib.rsCapability-isolated execution sandbox for running untrusted .obend specifications and evaluating emitted Activity<Plan, Response, Result> traces under resource/tick budgets.
obend-guicrates/obend-gui/src/main.rsInteractive native workbench (Obend.Workbench, lean/Obend/Workbench.lean) for inspecting live Slot readings (.usable with green support evidence vs. .blocked with named dead supports), C4 ancestry graphs, and Doc.Section pushout merges.
sandstorm/ & tests/sandstorm/, tests/run-obend-source.pyDifferential conformance suite (TEST_READY.md, /Users/ember/dev/obend/TEST_READY.md) holding Obend.Spec byte-for-byte compatible with minidregg's front-end test cohorts while testing 4,000 C4 DAG vectors and adversarial receipt/runner traces.

6. Complete Ledger of Closed Formal Lanes (0 sorryAx) vs. Open Architectural Decisions Awaiting Ember

6.1 Closed Formal Lanes (0 sorryAx in /Users/ember/dev/obend)

Every lane below is closed with 0 sorryAx in the Lean 4 kernel:

Lane IDCommit SHAPrimary ModulesHeadline Kernel Theorems (0 sorryAx)
LANE-EQUIREC89bd283Obend.Core.TypeGraph, RecursiveRows, EquiRecursiveAgreementagrees_iff, agrees_eq_candidate, alias_equivalent, full_annotated (flips full_corollary_fails)
LANE-HOLEd8db9f8Obend.Core.Term, Traverse, Holerename_fillHole, substitute_fillHole, typing_fillHole (CMTT explicit delayed substitution)
LANE-ELAB0c4a912Obend.Elab.Prepare, Coerce, Work, Gateprepare_checks, coerce_checks, 16-cell qle quantity matrix, checkWork_pos
LANE-C4d84c6c6Obend.Spec.C4, Proofs.C4, C4Vectorsc4_local_precedence, c4_monotonicity, c4_nodup, c4_suffix, OrderedPresentationInvariant
LANE-SPEC41e82a1Obend.Spec.Elaborate, Contract, LawDialaccepted_typed, accepted_never_refused, compose_typed, synthesizeLawRepair_sound
LANE-LINEARef8f814Obend.Core.LinearSoundness, Term.splitvalueStep_preserves_typing_and_linear1, value_consumes_linear, cbn_breaks_linear_occ
LANE-COST9b3d118Obend.Core.Cost, Clairvoyantmix_assoc_cost (CostWithin 8), self_field_no_cv, ticks_not_within_cv
LANE-SLOT / INTENTION7a19c04Obend.Slot, World, Selection, Obend.Intentionread_usable_supports, read_blocked_exact, permanent_blocked_forever, stale_iff_blocked, renew_repairs
LANE-JUDGE / LRAT5e82f10Obend.Judge, Lrat, Identity, CertifiedcheckBytes_sound, trivial_unregistrable, lrat_sound, noCarry_holds (215-step CaDiCaL LRAT in kernel)
LANE-FOLD / RECEIPT3c90e44Obend.Fold, Obend.Receipt, Merkle, CShake256receipt_sufficient, project_moves_run, recheck_sound, opens_binds, shape_strict, ceiling_checked, dial_invariant
LANE-LAW / KINDa4959c0Obend.Law, Std, Machine, Obend.Kindeffective_law_eval, inherited_never_widens, effective_refuses_first, ancestor_revised_stale, nominal_tooth
LANE-DOC6b21d09Obend.Doc, Patch, Merge, Section, Backlinkapply_compose, apply_invert, merge_isPushout, files_no_pushout, conflict_is_section, usable_unmoved
LANE-RUN2f77b81Obend.Run.Runner, Control, Shared, Durable, Fork, Reflectrefines, lost_reply_frozen, office_revoked_blocks_whileInOffice, joint_reachable_ticket_exclusive, exposed_once, stitch_is_commit, forged_descriptor_refused
LANE-LOWER / LEAFf0bb666Obend.Leaf, Obend.Lower.Tait, Bridgecompile_sound, strong_normalization_fixFree, lower_to_bend_sound
LANE-PARSIMONYf1e1406Obend.Slot (Just.lean & PolySlot.lean removed)Net -1,872 lines after tools/proofsize.py proved flat List Fact shorter than polynomial lattice

6.2 Open Architectural Decisions Awaiting Ember's Judgment

In /Users/ember/dev/obend-astra-decisions/astra/ASTRA-DECISIONS-20261101.md, /Users/ember/dev/obend-astra-decisions/review/DECISIONS-20261031.md, AstraDecisionEmbedded.lean, and AstraDecisionProbe.lean, six concrete architectural decisions are queued for Ember's final ruling:

  1. Decision 1 — Kernel Subtyping vs. Elaboration Coercions (Obend.Elab.Coerce vs. Polar TypeGraph):
  2. Current State: Obend.Core.Check uses symmetric equi-recursive agreement (EquiRecursiveAgreement.agrees), while width/depth/quantity subtyping is elaborated via explicit $\eta$-wrappers in Obend.Elab.Coerce (coerce_checks).
  3. Trade-off: Explicit $\eta$-wrappers keep the kernel small and respect the 16-cell quantity matrix, but allocate wrapper closures at runtime and cannot coerce inside invariant prototype S T specifications. Replacing sameType with directed polar simulation in TypeGraph (review/ALGEBRAIC-SUBTYPING-20261031.md) would make shareable width subtyping zero-cost at runtime.
  4. Decision 2 — Call-by-Need Heap Machine vs. Demand Machine (Obend.Core.Machine + Obend.Core.Clairvoyant):
  5. Current State: Obend.Core.Machine is a lazy call-by-name environment machine, and REPORT-COST-20261101.md proved that Hackett–Hutton whole-record clairvoyant CBV cannot type { a = 7, b = self.a }.b (self_field_no_cv).
  6. Decision Needed: To get proved memoization sharing across diamond self projections without losing lazy field convergence, should Core4 add a per-field lazy heap store (Launchbury / Sestoft mark-1 abstract machine with per-field thunks) and prove representation adequacy against Obend.Core.Term?
  7. Decision 3 — Binding Export.verify to Authenticated World Roots (AstraDecisionProbe.lean):
  8. Current State: As demonstrated in AstraDecisionProbe.lean, Obend.Receipt.Export.verify validates a receipt against a caller-supplied dated Merkle root.
  9. Decision Needed: Should Export.verify require an authenticated Ledger.CommitRoot (from Obend.Run.Shared / Obend.Vault) or principal signature so unanchored receipts cannot be verified against self-signed roots?
  10. Decision 4 — Deep Embedding vs. Pruning of the Late Nov 1 Domain Leaves (ASTRA-COUNCIL-20261101.md):
  11. Current State: The 300-module spine at checkpoint-20261101 (6266457) is deeply connected end-to-end, whereas ~120 modules added in the late Nov 1 burst (Compat, Vcs, Verify, Effects, Capabilities, Consensus) are self-contained shallow models.
  12. Decision Needed: Should those late domain leaves be refactored to compile through Obend.Spec / Obend.Core.Term / Obend.Run (as AstraDecisionEmbedded.lean demonstrates by running a parsed .obend specification through Spec.FrontEnd.accept → Law.effectiveLaw → Run.Runner), or pruned back to checkpoint-20261101 + LinearSoundness + Lower + Kind + Quote?
  13. Decision 5 — Reconciling DESIGN.md §1 with Core4's Actual Type Grammar (FABLE-COUNCIL-20261101.md):
  14. Current State: DESIGN.md §1 still lists dependent pi, sigma, id, refl, jElim constructors from an early draft, whereas all theorems in Obend.Core.* are proved over Core4 (Ty.arrow, Ty.field, Ty.specification, Ty.prototype, Ty.custody, Ty.variant, Ty.computation) and dependent proof checking is delegated to Obend.Judge certificates.
  15. Decision Needed: Update DESIGN.md §1 so the prose matches the proved Core4 + Obend.Judge architecture (or decide whether any indexed dependent fragment belongs inside Core.Ty).
  16. Decision 6 — Strict CBV vs. Lazy CBN for Linear Capabilities (Obend.Core.LinearSoundness):
  17. Current State: LinearSoundness.lean proves exact single-consumption (occ 0 = 1 → 0) under strict ValueStep for .linear arrows (app_linear_val) and .split, and proves cbn_breaks_linear_occ for lazy CBN substitution.
  18. Decision Needed: Should Obend.Core.Machine (the primary demand machine) switch .linear-annotated applications to strict call-by-value evaluation so the main machine inherits LinearSoundness directly?
Dossier 23 · Direct Source Re-Scout: `minidregg-lanes` (315 Entries), `claudesplosion` (67 Lanes), `nanodregg-scout` (27 Scouts), `delvetalk2-lanes` (66 Entries), `drorb` & `DreggCloud` (110 KB)
Working field notes & source index below, with theorem anchors, code pointers, git history, and open questions.

Dossier 23: The Capability OS, Semantic Kernel, and Sovereign Edge Firmament (minidregg, claudesplosion, nanodregg, delvetalk2, drorb, DreggNet, DreggCloud)

[!IMPORTANT] Primary-Source Archaeological Scope: Every claim, architectural invariant, theorem signature, and worktree disposition in this monograph is excavated directly from Ember's primary repositories and worktree farms on disk as of November 2, 2026: - /Users/ember/dev/minidregg (8,953 commits at Nov 2 snapshot c4ecabe0f; actively developed through Nov 14, 2026 at origin/main c60818647 + append-growth 5e05cf4fc, with 315 lane entries in /Users/ember/dev/minidregg-lanes — see Dossier 28 §4.2), /Users/ember/dev/minidregg-wt (3 git worktrees + seats/), and /Users/ember/dev/claudesplosion (160 commits, HEAD 6c45f2e, 2026-10-03). - /Users/ember/dev/nanodregg (871 commits, HEAD 80c622a, 2026-10-27), /Users/ember/dev/nanodregg-scout (27 scout directories C1..T2), and /Users/ember/dev/nanodregg-wt (9 git worktrees). - /Users/ember/dev/delvetalk (21 commits, HEAD 9f70515) & /Users/ember/dev/delvetalk2 (2,583 commits, HEAD fb2179968, 2026-10-15) & /Users/ember/dev/delvetalk2-lanes (66 lane entries: 45 git worktrees + 21 handoff/artifact files). - /Users/ember/dev/drorb (618 commits, HEAD ec01e68, 2026-10-06), /Users/ember/dev/drorb-swarm-salvage (7 salvaged Lean 4 modules + 50.6 KB patch), /Users/ember/dev/DreggNet (520 commits, HEAD 2c574b6, 2026-08-06), and /Users/ember/dev/DreggCloud (88 commits, HEAD e9ff17a, 2026-08-29).


1. Executive Thesis & Architectural Lineage

Across more than 13,800 commits and 420+ parallel worktree lanes, Ember constructed a vertically integrated, formally verified capability operating system, semantic kernel, object-capability social world, post-quantum edge network, and sovereign application cloud. The central thesis uniting these repositories is Zero-Twin Semantic Ownership: Lean 4 owns the meaning, the wire format, the transition semantics, the capability admission laws, and the soundness proofs; Rust and Pancake/C move bytes, schedule I/O rings, and enforce OS sandbox boundaries without ever deciding semantic admissibility.

1.1 The Four Structural Movements

  1. From Hand-Written Twins (breadstuffs) to the Derived Semantic Computer (minidregg & claudesplosion):
  2. In /Users/ember/dev/minidregg/ATLAS.md, Ember diagnosed the fundamental pathology of /Users/ember/dev/breadstuffs (~2.89M LOC: 1.82M Rust across 137 crates, 913K Lean across 1,039 files, 160K TS/JS): every derivation compiler had been built and proved at $n = 1$, and then a hand-written twin was maintained beside it in production.
  3. minidregg ("Mini") was founded on a strict ≤10% LOC mandate where the derived path is the only path: a single Lean-compiled host binary serves one append-only state-and-event log (World), evaluates total capability programs written in Objective Bend, enforces relational predicates (Pred), and emits commitment proofs over the Selvage BaseFold + sumcheck + Poseidon2/Goldilocks stack.
  4. Between September 29 and October 3, 2026, Ember orchestrated /Users/ember/dev/claudesplosion: first a 23-lane + 6-lane adversarial formal audit (01..23, mini-1..6, 99-SYNTHESIS.md) that exposed 22+ confirmed-at-source defects across breadstuffs, fn, and minidregg, and then a 67-lane subagent takeover (planning/handoff-2026-10-02/LANE-INDEX.md) that hardened Mini's resident agent loop (Hermes), Sandstorm application marketplace, DREGG/RESOURCE/TRANSACTION v9 compute funding, and prospective document privacy (OBSERVE-CHALLENGE/v9 + OBSERVE-SIGNED/v9).
  5. The Spec-First Single-Writer Distillation (nanodregg):
  6. On October 16, 2026, drawing on the lessons of minidregg's 273 lanes, Ember launched /Users/ember/dev/nanodregg under a 12-law CHARTER.md and a monotonic SIZE.md ratchet: 32,077 deployed Lean lines, 6,703 named theorems, 0 partial defs, 0 sorry, 99 lines of Rust in nd, and 164 lines of Rust in nd-supervise, backed by 27 architectural scouts (nanodregg-scout C1..T2). Every top-level claim in Spec/README.md (OneHistory, Integrity, Consent, Shown) is simultaneously proved of the deployed system (top_*) and refuted on a lazy trivial counterparty (floor_*).
  7. The Live Object-Capability Social Town (delvetalk & delvetalk2):
  8. In /Users/ember/dev/delvetalk2 (2,583 commits, live at gsb.fg-goose.online for delve.town and obend.fg-goose.online), Ember turned the Objective Bend kernel into a live multi-world object town (Record, Thing, Kind, Avatar, Library, Mending, Policy, Tide, Workshop) with Faré-style prototype inheritance (fix/mix), Spween scene graphs (world/station.spween), unified law / rule / want modal clauses (ObjectiveBendLaw.lean), stroke-verified LLM "enchantment" painting (Host/Enchant.lean), cross-edition world carry (Host/Carry.lean), and native AT Protocol (did:web, signed k256 MST + DAG-CBOR, 18 online.fg-goose.gsb.* lexicons) federation.
  9. The Verified Post-Quantum Edge, Execution Mesh, and Sovereign Cloud (drorb, DreggNet, DreggCloud):
  10. At the network and hosting perimeter, /Users/ember/dev/drorb provides a formally verified post-quantum (ML-KEM-768 + ML-DSA-65 + verified FIPS 202 Keccak) reverse proxy, TLS 1.3 / HTTP/2 / HTTP/3 / QUIC v1 terminator, and full Tailscale (ts2021) control and data plane in Lean 4 + HOL4 Pancake (datacake) + Rust io_uring/kqueue.
  11. /Users/ember/dev/DreggNet pairs a Lean 4 Pancake reactor compiler (Dsl/Reactor.lean, Dsl/EmitPancake.lean) with a 6-crate Rust execution cloud (dreggnet-exec wasmi, dreggnet-bridge, dreggnet-control boringtun WireGuard mesh, dreggnet-gateway).
  12. /Users/ember/dev/DreggCloud binds the entire stack into a House / World / Room sovereign custody firmament (dreggcloud-house), featuring hybrid Ed25519 + ML-DSA (cipherclerk) command envelopes, DEOS Shadow DOM view trees (dc-delivery-*), live MPC-TLS (TLSNotary schema v14), exact Universe world custody (universe-house-custody schema v15), prepaid service archive custody (service-house-custody schema v16), and the grain-serve Sandstorm-serve capability weld.
flowchart TD
    subgraph Edge["Verified PQ Edge & Mesh (drorb + DreggNet)"]
        DRORB["drorb (618 commits)\nVerified TLS 1.3 / H2 / H3 / QUIC v1\nML-KEM-768 + ML-DSA-65 + Keccak\nTailscale ts2021 Control & Data Plane"]
        SALVAGE["drorb-swarm-salvage\nQuicServer.lean (1,786L) + Framing.lean\nDeployPlus / CacheControl / RetryAfter"]
        DNET["DreggNet (520 commits)\nLean 4 Reactor DSL -> Pancake C\nwasmi Exec + WireGuard Mesh + Bridge"]
        DRORB -.->|"un-inerts braids 6..10"| SALVAGE
    end

    subgraph Cloud["Sovereign Custody Firmament (DreggCloud)"]
        HOUSE["DreggCloud (88 commits)\nHouse / World / Room Ontology\nCipherclerk Ed25519 + ML-DSA"]
        SHC["service-house-custody (Schema v16)\nuniverse-house-custody (Schema v15)\nTLSNotary MPC-TLS (Schema v14)"]
        GSERVE["grain-serve\nsandstorm-serve Capability Weld\nBearerGate + AssuranceRung"]
        HOUSE --> SHC
        HOUSE --> GSERVE
    end

    subgraph Kernels["Semantic Kernels & Capability OS (minidregg + nanodregg)"]
        MINI["minidregg (8,953 commits)\nKernel/World.lean + Objective Bend Core4\nSelvage BaseFold + Poseidon2 + Host"]
        MLANES["minidregg-lanes (273 entries)\n140 merged clean | 67 merged dirty | 40 unmerged\n+ minidregg-wt (3 worktrees)"]
        CLAUDE["claudesplosion (160 commits)\n23+6 Review Lanes (99-SYNTHESIS.md)\n67-Lane Deep Takeover + v9 Contracts"]
        NANO["nanodregg (871 commits)\n32,077L Lean | 6,703 Theorems | 0 sorry\n27 Scouts (C1..T2) + 9 Worktrees"]
        MINI <--> MLANES
        CLAUDE -->|"audited & handed off"| MINI
        MINI -->|"distilled into single-writer spec"| NANO
    end

    subgraph Town["Live Object-Capability Town (delvetalk + delvetalk2)"]
        DT2["delvetalk2 (2,583 commits, Edition 20)\ngsb.fg-goose.online (delve.town)\nRecord / Thing / Kind + Spween"]
        DT2LAW["ObjectiveBendLaw.lean (law/rule/want)\nHost/Enchant.lean (Stroke-Checked LLM)\nHost/Carry.lean + ATProto MST/DAG-CBOR"]
        DT2LANES["delvetalk2-lanes (66 entries)\n44 merged clean | 1 unmerged (perf) | 21 handoffs"]
        DT2 --> DT2LAW
        DT2 <--> DT2LANES
    end

    DRORB --> HOUSE
    DNET --> HOUSE
    GSERVE --> MINI
    MINI --> DT2

2. /Users/ember/dev/minidregg (8,953 Commits): The Lean-Owned Semantic Computer

2.1 The 17 Atlas Laws and the Diagnosis of breadstuffs

In /Users/ember/dev/minidregg/ATLAS.md, Ember documented why minidregg exists. While breadstuffs proved that logic-arithmetization compilers, protocol state-machine generators, and Lean-to-Rust codegen worked individually, each compiler in breadstuffs had been deployed at $n = 1$ alongside a hand-written twin:

  • lean/Proto/Specs/Noise.lean compiled only Noise; the other 7 protocols in crates/proto/ were hand-written Rust.
  • lean/Drakov/Circuits/ByteDecompCircuit.lean discharged only 1 of 20 circuit families; the other 19 were hand-written Plonky2 gates in crates/circuit/.
  • lean/Contracts/Generated/ emitted only 3 files (poseidon_constants.rs, transcript_schedule.rs, error_codes.rs), leaving 114 hand-written files in crates/core/.

ATLAS.md establishes 17 Design Laws for minidregg:

  1. Law 1 (Total & Safe by Construction): User programs are written in Objective Bend, a total functional/imperative capability language with affine linear resources and bounded recursion (fold/bend). Every well-typed program terminates and has a deterministic static polynomial cost bound.
  2. Law 2 (Automatic Arithmetization): The user never writes circuits, AIR rows, or witness generators. Compiler/TraceTable.lean lowers total programs directly to Rank-1 / Degree-2 trace tables over Goldilocks ($\mathbb{F}_p, p = 2^{64} - 2^{32} + 1$).
  3. Law 3 (Automatic Incrementalization / IVC): State machines S → I → S automatically compile to incremental steps without user-managed folding transcripts.
  4. Law 4 (Automatic Privacy / Oblivious Lowering): Values carry explicit security labels; branches on secret data lower to branchless MUX rows (SUMS-DESIGN.txt).
  5. Law 5 (Capability-Safe Shared State): Shared state is factored into independent Cells (CellId → Cell R); transactions declare their cell footprint (read, write, claim, retire) upfront.
  6. Law 6 (Content-Addressed Code & Data): Every program, type, and cell value is addressed by its canonical cryptographic digest.
  7. Law 7 (One Cryptographic Primitive — Poseidon2 over Goldilocks): Hashing, Fiat-Shamir transcripts, Merkle commitments, and AEAD stream encryption all derive from a single permutation (Selvage/Poseidon2Spec.lean).
  8. Law 8 (One Proof System — BaseFold + Sumcheck): Multilinear polynomial commitments via foldable Reed-Solomon codes (Selvage/BaseFold*) + sumcheck replace the 10-way proof-system zoo (Plonky2, Halo2, Nova, SuperNova, Groth16, Stark, KZG, Bulletproofs) of breadstuffs.
  9. Law 9 (Lean Owns the Meaning; Rust Owns Only Execution & OS Sandboxing).
  10. Law 10 (No Hand-Written Twins).
  11. Law 11 (Everything Speaks One Self-Describing Schema): Pred / Kernel.Value replaces separate SQLite, Protobuf, JSON, and custom binary wire schemas.
  12. Law 12 (Effects Are Capabilities over the Same Log): HTTP, timers, LLM calls (Hermes), and Sandstorm grain bridges are algebraic effects (perform) resolved as journal turns.
  13. Law 13 (Local-First, Sync-by-Merkle-Diff).
  14. Law 14 (Upgrades Are Migration Functions).
  15. Law 15 (Errors Carry Blame): Refusals are typed data values naming the exact violated precondition or law clause.
  16. Law 16 (Zero Configuration): One host binary, one store directory.
  17. Law 17 (Never Slower Over Time — $O(\log n)$): Added in late October 2026 (ATLAS.md) after profiling live stores: no admitted operation or read may scale linearly with journal height $H$ or total cell count $N$; every index must be maintained incrementally or backed by a persistent balanced tree (SparseMap / binary Merkle trie).

2.2 Maturity Classification (S / A / P / D / B) in PROJECT.md

To prevent documentation inflation, /Users/ember/dev/minidregg/PROJECT.md grades every subsystem with an explicit 5-tag maturity code:

  • [S] Spec: Total Lean definitions.
  • [A] Axiom-clean: #print axioms verified clean (only propext, Quot.sound, Classical.choice where permitted, zero sorry) via Assurance/TokenCensus.lean.
  • [P] Mechanized Proof: Non-vacuous soundness/refinement theorem discharged in Lean 4.
  • [D] Derived: Emitted automatically from Lean into Rust/C/tables without hand-written glue.
  • [B] Bridged: Wired into the live host binary and exercised by end-to-end executable tests.

2.3 The Semantic Kernel: Kernel/World.lean and Kernel/HostRefinesWorldCore.lean

At the heart of minidregg is /Users/ember/dev/minidregg/Kernel/World.lean. The entire machine state World R is a single sparse map from CellId (Nat) to Cell R, where cell 0 is a distinguished system cell (sysCellId) partitioned by SysSpace:

lean
-- /Users/ember/dev/minidregg/Kernel/World.lean:38-115
/-- A single cell in the world state.
    - `program`: digest of the admission predicate `P : R → R → R → Bool`.
      The predicate takes `(oldValue, newValue, txContext)` and returns `true`
      iff the transition is permitted.
    - `value`: the current typed value (`R`).
    - `nonce`: monotone counter incremented on every write.
    - `retired`: once `true`, the cell is permanently frozen / consumed
      (used for one-shot capabilities and nullifiers). -/
structure Cell (R : Type) where
  program : Nat
  value   : R
  nonce   : Nat
  retired : Bool
  deriving DecidableEq, Repr

/-- The world state is a sparse map from `CellId` to `Cell R`. -/
abbrev World (R : Type) := Minidregg.Theory.SparseMap CellId (fun _ => Cell R)

/-- Read-only ROM commitment view used by ZK / IVC step circuits. -/
structure RomOnly where
  root : Nat
  deriving DecidableEq, Repr

/-- Sub-namespaces inside the distinguished system cell (`sysCellId = 0`). -/
inductive SysSpace where
  | journal    -- append-only turn journal
  | head       -- current chain head digest / height
  | retired    -- nullifier / retired-cell set
  | parent     -- parent cell provenance
  | spent      -- spent capability tokens
  | allowance  -- storage / compute quota accounting
  deriving DecidableEq, Repr

/Users/ember/dev/minidregg/Kernel/HostRefinesWorldCore.lean connects the imperative Host loop (HostState, Host.submit) to the pure relational World.step transition:

lean
-- /Users/ember/dev/minidregg/Kernel/HostRefinesWorldCore.lean:45-168
/-- Case analysis for any cell across a valid `World.step`: a cell is either
    untouched, freshly created, updated in place with `nonce' = nonce + 1`,
    retired, or is the system cell recording the journal turn. -/
theorem step_cell_cases
    {w w' : World Value} {tx : Transaction}
    (hstep : World.step w tx = .ok w') (cid : CellId) :
    StepCellOutcome w w' tx cid

/-- Rewriting an append-only projection without a valid journal turn is impossible. -/
theorem appendOnly_rewrite_has_no_turn
    {w w' : World Value} (h : IsAppendOnlyRewrite w w') :
    ∀ tx, World.step w tx ≠ .ok w'

/-- Every confirmed receipt returned by the Host represents an admitted World state. -/
theorem confirmed_represents
    {s s' : HostState} {req : SubmitReq} {resp : SubmitResp}
    (h : Host.submit s req = (s', .confirmed resp)) :
    Represents s'.store resp.worldRoot

/-- T4 refinement target: a successful `Host.submit` corresponds to a valid `World.step`
    from the represented pre-state to the represented post-state. -/
theorem host_submit_is_step
    {s s' : HostState} {req : SubmitReq} {resp : SubmitResp}
    (hinv : HostInvariant s)
    (hsub : Host.submit s req = (s', .confirmed resp)) :
    ∃ w w', Represents s.store w ∧ Represents s'.store w' ∧
      World.step w req.toTransaction = .ok w'

2.4 Objective Bend Core4: PROOF-ENG-DESIGN.md, SUMS-DESIGN.txt, and EVENTS-DESIGN.txt

Three foundational design documents in /Users/ember/dev/minidregg/ record how Objective Bend's operational semantics and proof engineering were scaled to handle real applications without proof-Checking blowups:

  1. Structural Case-Principle Refactoring (/Users/ember/dev/minidregg/PROOF-ENG-DESIGN.md):
  2. Prior to the Core4 refactor, Theory/ObjectiveBendTyping.lean had 14 positional PartialTyping.rec inductions (each with 29 positional arguments fun _ _ _ _ ... =>) and 5 duplicated stepRaw case trees across 2,400 lines. Adding a single typing constructor broke 406 positional binders.
  3. Ember introduced Theory/ObjectiveBendStepCases.lean, packaging the per-step case split into a single reusable structure StepCases (P : State → Prop) with one named field per semantic reduction family (stepVar, stepLet, stepPair, stepRecord, stepSum, stepPerform, etc.) and split the subproofs into Theory/ObjectiveBend/Cases/{Activities,Arithmetic,Functions,Pairs,Records,Sums}.lean.
  1. Exhaustive Closed Variant Rows & Oblivious Sums (/Users/ember/dev/minidregg/SUMS-DESIGN.txt):
  2. Implemented in lane W1.3, SUMS-DESIGN.txt added tagged union types Ty.variant (row : Ty) ( reusing Ty.rowEmpty and Ty.rowExtend label payload rest from records) along with Term.inject (label : Label) (payload : Term), Term.case (scrutinee : Term) (branches : Branches), and Term.ifBool:
lean
   -- /Users/ember/dev/minidregg/SUMS-DESIGN.txt:28-64
   -- Typing rules for closed row-typed variants:
   --   RowHas label A row
   --   ----------------------------------------------- (T-Inject)
   --   Γ ⊢ Term.inject label e : Ty.variant row
   --
   --   Γ ⊢ s : Ty.variant row
   --   BranchesCover row branches C
   --   ----------------------------------------------- (T-Case)
   --   Γ ⊢ Term.case s branches : C
  • Crucially, Ty.variant row is shareable iff every payload type in row is shareable (Shareable (Ty.variant row) ↔ RowAllShareable row), and in the oblivious circuit backend (Law 4), Term.case over a secret tag lowers to evaluating all branches over a canonical dummy payload of each branch's type and selecting the result via a chain of Primitive.labelEqual multiplexers (MUX).
  1. Quiescent Algebraic Activities (/Users/ember/dev/minidregg/EVENTS-DESIGN.txt):
  2. Implemented in lane W3.EVENTS, EVENTS-DESIGN.txt replaced ad-hoc host callbacks with a typed algebraic effect discipline:
  3. Types: Ty.computation (plan response result : Ty) (written Activity<Plan, Response, Result> in surface syntax).
  4. Terms: Term.perform (plan : Term) and Term.done (value : Term), sequenced by let x <- e1; e2.
  5. The Quiescent Yield Invariant: By making Ty.computation non-shareable (Shareable (Ty.computation _ _ _) = False), a computation can never be stored inside a shared heap cell or evaluated underneath a lazy update frame on the heap! Consequently, when stepRaw reaches Term.perform vPlan, the machine is guaranteed to have zero Frame.update frames on its continuation stack (no_update_frame_at_yield). The suspended continuation K is a pure first-order value that is serialized into the cell's journal turn, and the turn boundary acts as Nakano's later modality ($\triangleright$)—preserving total normalization within every single turn while permitting arbitrary multi-turn reactive protocols across turns.

2.5 Cross-Repository Cryptographic & Wire Contracts (CROSS_REPO_CONTRACTS.md)

/Users/ember/dev/minidregg/CROSS_REPO_CONTRACTS.md freezes the exact byte-level and field-level contracts shared between minidregg and external verification consumers:

  • UwueavePreoProjectionV2 / ProjectionV3: Deterministic pre-order and projection artifact schemas emitted into prover/generated/uwueave_preo_projection_v2.rs.
  • RuntimeAuthV4: Capability-bound authentication envelope binding (principalPk, worldRoot, cellId, nonce, expiryHeight, actionDigest).
  • ZkmlSuiteArtifact & BaseFoldBcs*: Fixes the Goldilocks Poseidon2 sponge schedule (rate = 8, capacity = 4, domain tag in capacity lane 0), BaseFold Reed-Solomon folding arity, and Rank-1 MLE sumcheck transcript ordering so Lean verifiers (Selvage/) and Rust provers (prover/) agree bit-for-bit.

3. /Users/ember/dev/claudesplosion (160 Commits): The 23+6 Review Swarm & 67-Lane Deep Takeover

/Users/ember/dev/claudesplosion (160 commits by Ember between September 29 and October 3, 2026, HEAD 6c45f2e) is the forensic nerve center where Ember ran two massive waves of formal audit and subagent orchestration:

  1. The 23+6 Read-Only Adversarial Audit Swarm (00-README.md, 99-SYNTHESIS.md, 01-crypto.md .. 23-fn-assurance-tests.md, mini-1-kernel.md .. mini-6-docs-claims-evidence.md).
  2. The 67-Lane Deep Mini Takeover & Codex Handoff (planning/handoff-2026-10-02/LANE-INDEX.md, planning/handoff-2026-10-03/START-CODEX.md, planning/docuverse-convergence-2026-10-02.md, planning/final/).

3.1 Round 1 & Round 2 Adversarial Audit (99-SYNTHESIS.md & mini-1..6)

In /Users/ember/dev/claudesplosion/99-SYNTHESIS.md, Ember synthesized 22+ confirmed-at-source vulnerabilities and proof vacuities across breadstuffs and fn, alongside the 6 deep Opus lanes (mini-1 through mini-6) auditing minidregg at commit abe988d:

ID / LaneTarget File & LineConfirmed-at-Source FindingSeverity / Impact
F-01 (01-crypto, 02-circuit)crates/proof/src/rotated_replay.rs:284-319Tail public-input anchors in rotated IVC replay were compared against self-supplied witness slots rather than the caller's pinned anchor digest.Critical: Forged continuation tail accepted by replay verifier.
F-02 (06-platform, 07-runtime)crates/platform/src/sev_snp.rs:142-188AMD SEV-SNP attestation report parser verified the Ask/Vcek signature chain but ignored the 64-bit guest_policy bitfield (allowing debug-mode guests if measurement matched).High: Debug-policy guest admitted under hardware attestation.
F-03 (10-lean-compiler)lean/Compiler/VcgSoundness.lean:91-134vcg_run_sound quantified over (h : ∀ x, RecChained x) where RecChained was uninhabited for non-empty loops, making the soundness theorem vacuously true on loops.High: Vacuous Lean 4 theorem premise.
F-04 (09-lean-drakov)lean/Drakov/BatchVerify.lean:64-82verifyBatch took (_vk : VerifyingKey) with an underscore-prefixed unused binder, checking transcript consistency without binding the verifying key.High: Spec-level key-binding omission.
F-05 (11-lean-proto-consensus)lean/Consensus/GstLiveness.lean:48-71World.gst_liveness included the conclusion inside the WellFormedSchedule hypothesis bundle.Medium: Tautological liveness theorem.
F-06 (20-fn-id-cred)fn/crates/cred/src/digest.rs:53Credential revocation digest truncated SHA-256 to 64 bits (u64::from_le_bytes(h[0..8])), reducing collision resistance to $2^{32}$.High: Birthday-bound credential collision.
mini-1-kernelminidregg/Kernel/World.leanAudited World.admit and Kernel/ at abe988d; verified zero sorry in Kernel/ and identified 4 untracked frontier files (ApplicationDispatchAgentPurse, ApplicationDispatchAgentReserve, FnConsumerFrontierProofs, FnConsumerFrontierScenario).Clean core; frontier isolated.
mini-2-compiler-theoryminidregg/Theory/ & Compiler/Audited Objective Bend typing/step proofs and the Loom → Selvage rename across Compiler.lean.Verified total step proofs.
mini-3-selvage-assurance-hostminidregg/Selvage/ (149 files, 76,553 lines)Audited FRI/RS BaseFold proximity gaps, sumcheck soundness, and Assurance/TokenCensus.lean.76.5K lines of axiom-clean proof math.
mini-4-nativeminidregg/native/spk-host/src/sandbox.rsAudited bwrap_args, open_protected_directory, verify_var_volume, verify_executable, and spawn_sandbox.Confirmed strict bubblewrap confinement.
mini-5-prover-scripts-gatesminidregg/prover/src/sumcheck.rsDocumented that minidregg-prover (prove_sumcheck/verify_sumcheck + rank1.rs) was at that commit a degree-1 MLE sumcheck engine over sha3, not yet wired to the full Selvage BaseFold BCS openings.Triggered the October BaseFold BCS bridge lanes (wave-assurance-detwin).
mini-6-docs-claims-evidenceminidregg/docs/Cross-checked all 12 top-level claim documents against git tree abe988d by SHA-256 and import-graph walk.Established the PROJECT.md [S][A][P][D][B] truth tags.

3.2 Commit ce2712c: The 67-Lane Deep Mini Takeover (LANE-INDEX.md)

On October 2, 2026, in commit ce2712c ("Document deep Mini takeover with 67 lane dispositions and artifact boundaries"), Ember recovered and cataloged all 67 parallel subagent lanes spawned under root session 01a0fae5-8eb3-7392-83ee-4179c7664edc inside /Users/ember/dev/claudesplosion/planning/handoff-2026-10-02/. The takeover organized the 67 lanes across 11 domain handoff dossiers:

  1. CLEANUP.md: Repo hygiene, dead-twin deletion, and warning zeroing.
  2. DOCUMENTS.md: Documentverse (createDocument, editElement, createAtom, editAtom), prospective encryption, and observation union v9.
  3. EXPERIENCE-ROOMS.md: Workrooms, channels, collaborative cursors, and room templates.
  4. INTEGRATION.md: Capsule sealing (family-90562d76), 24-role manifest, and cross-lane merge order.
  5. ORCHESTRATOR.md: Multi-agent seat scheduling and lane worktree isolation.
  6. PLATFORM-PAY.md: DREGG/RESOURCE/TRANSACTION v9, agent purses, provider settlement, and compute-funding contracts.
  7. PROGRAMMING-PERFORMANCE.md: Hot-path caching, proof trimming, and room new latency reduction.
  8. RESIDENT.md: Hermes autonomous resident loop, tool dispatch, and controller write scopes.
  9. REVIEW-STRANDS.md: Formal proof closure and TokenCensus enforcement.
  10. SERVICE.md: HTTP door, WebSocket push, and multi-host replication.
  11. SPK.md: Sandstorm package (spk-host) confinement, app market catalog, and grain bridges.

Below is the complete 67-lane disposition ledger excavated from /Users/ember/dev/claudesplosion/planning/handoff-2026-10-02/LANE-INDEX.md:

#Lane NameSession IDDomain DossierDisposition in LANE-INDEX.md
1abuse-caps01a10042SERVICE.mdLanded: Rate-limit & quota cells enforced on public doors.
2agent-purse01a10088PLATFORM-PAY.mdLanded: Per-agent bounded spending purse cell & kernel predicate.
3agent-reserve01a10091PLATFORM-PAY.mdLanded: Escrow reserve & settlement release for autonomous turns.
4app-batch-101a10104SPK.mdLanded: First batch of Sandstorm .spk grain adapters & bridges.
5app-batch-201a10109SPK.mdLanded: Second batch of .spk grain adapters (Etherpad/Wekan/Davros).
6app-batch-301a10115SPK.mdLanded: Third batch of .spk grain adapters (GitWeb/Radicale/Collections).
7app-batch-401a10121SPK.mdLanded: Fourth batch of .spk grain adapters + capability bridge tests.
8app-market01a10130SPK.mdLanded: Content-addressed .spk catalog cell & install flow.
9app-market-ui01a10136SPK.mdLanded: Browser UI for one-click grain launch from market catalog.
10ash-furnace01a10144CLEANUP.mdLanded: Burned dead Rust/Lean shims left from pre-Selvage migration.
11assurance-census01a10152REVIEW-STRANDS.mdLanded: Assurance/TokenCensus.lean zero-sorry / zero-axiom gate.
12auth-v4-bind01a10160SERVICE.mdLanded: RuntimeAuthV4 request-to-worldRoot signature binding.
13bcs-bridge01a10168REVIEW-STRANDS.mdLanded: BaseFold BCS transcript schedule alignment between Lean & Rust.
14book-conservation01a10175PLATFORM-PAY.mdLanded: Double-entry balance conservation theorem across Book turns.
15bwrap-hardening01a10183SPK.mdLanded: native/spk-host/src/sandbox.rs mount-namespace & fd-hygiene gates.
16capsule-seal01a10190INTEGRATION.mdLanded: Sealed 24-role runtime capsule family-90562d76.
17carry-public568801a10198INTEGRATION.mdLanded: Historical Public5688 IMAGE/v1 carry lift into Mini.
18cell-sparse-map01a10205PROGRAMMING-PERFORMANCE.mdLanded: Persistent radix-tree SparseMap replacing association lists.
19channel-backscroll01a10212EXPERIENCE-ROOMS.mdLanded: Windowed cursor pagination for room channel transcripts.
20client-sdk-ts01a10220SERVICE.mdLanded: Browser/Node client SDK speaking canonical SubmitReq frames.
21codex-handoff-prep01a10228ORCHESTRATOR.mdLanded: START-CODEX.md baseline verification & diagnostic scripts.
22compute-funding-v901a10235PLATFORM-PAY.mdLanded: DREGG/RESOURCE/TRANSACTION v9 tag 6 computeFunding.
23content-privacy01a10242DOCUMENTS.mdLanded: content_privacy.rs action classifier & plaintext leak refusal.
24controller-write-scopes01a10250RESIDENT.mdLanded (c9d91bed): Scoped write capabilities for autonomous controllers.
25cursor-presence01a10258EXPERIENCE-ROOMS.mdLanded: Ephemeral presence & cursor broadcast over room WebSockets.
26dead-code-reap01a10265CLEANUP.mdLanded: Pruned 18,400 lines of unreachable legacy adapter code.
27doc-atoms-crdt01a10272DOCUMENTS.mdLanded: Causal tree / atom addressing for collaborative documents.
28doc-export-md01a10280DOCUMENTS.mdLanded: Deterministic Markdown projection from document atom cells.
29docuverse-v901a10288DOCUMENTS.mdLanded: OBSERVE-CHALLENGE/v9 + OBSERVE-SIGNED/v9 union contract.
30door-http-etag01a10295SERVICE.mdLanded: World-root-bound HTTP ETag and 304 Not Modified short-circuit.
31effects-activity01a10302REVIEW-STRANDS.mdLanded: EVENTS-DESIGN.txt Ty.computation quiescent yield proofs.
32epoch-roster-kem01a10310DOCUMENTS.mdLanded: protected_object/ epoch+roster-bound symmetric key wrapping.
33evidence-frontier01a10318REVIEW-STRANDS.mdPreserved WIP: FnConsumerFrontierProofs & FnGatewayPolicy.lean.
34fast-hash-native01a10325PROGRAMMING-PERFORMANCE.mdLanded: Vectorized Goldilocks Poseidon2 FFI bridge with parity check.
35fn-e2-cache01a10332PROGRAMMING-PERFORMANCE.mdPreserved in minidregg-wt: Materialized projection cache worktree.
36grain-bridge-http01a10340SPK.mdLanded: Capability-gated HTTP reverse proxy into running .spk grains.
37grain-volume-seal01a10348SPK.mdLanded: Content-addressed snapshot/restore of grain /var volumes.
38hermes-context-pack01a10355RESIDENT.mdLanded: Deterministic token-budgeted room context packer for Hermes.
39hermes-live01a10362RESIDENT.mdActive Frontier (+24 commits): End-to-end live model turn loop.
40hermes-tool-sandbox01a10370RESIDENT.mdLanded: Capability-attenuated tool execution for Hermes turns.
41host-refines-world01a10378REVIEW-STRANDS.mdLanded: Kernel/HostRefinesWorldCore.lean step_cell_cases & confirmed_represents.
42host-submit-batch01a10385PROGRAMMING-PERFORMANCE.mdLanded: Group-commit batch admission in Host to amortize fsync.
43invite-capabilities01a10392EXPERIENCE-ROOMS.mdLanded: Single-use and bounded-use room capability invite links.
44iter-speed01a10400PROGRAMMING-PERFORMANCE.mdActive Frontier (+16 commits): Incremental Lake/Cargo build acceleration.
45journal-segment-crc01a10408SERVICE.mdLanded: Frame-level Poseidon2 MAC + length prefix on journal segments.
46key-custody-broker01a10415RESIDENT.mdLanded: Out-of-process API key custody so model keys never enter grain RAM.
47law-blame-diagnostics01a10422REVIEW-STRANDS.mdLanded: Structured refusal blame trees returned on rejected turns.
48media-blob-cas01a10430EXPERIENCE-ROOMS.mdLanded: Content-addressed attachment store with cell-gated read tokens.
49merkle-proof-trim01a10438PROGRAMMING-PERFORMANCE.mdLanded: Pruned redundant sibling hashes in multi-cell read proofs.
50migration-carry-v101a10445INTEGRATION.mdLanded: Schema migration combinators for cell upgrades (Law 14).
51obend-sums01a10452REVIEW-STRANDS.mdLanded: SUMS-DESIGN.txt closed variant rows & oblivious MUX lowering.
52obend-step-cases01a10460REVIEW-STRANDS.mdLanded: PROOF-ENG-DESIGN.md StepCases modular proof architecture.
53oracle-v7-qual01a10468PLATFORM-PAY.mdActive Frontier (+23 commits): External price/compute oracle qualification.
54pay-settlement-receipt01a10475PLATFORM-PAY.mdLanded: Verifiable provider settlement receipts bound to turn digests.
55perf-apps01a10482PROGRAMMING-PERFORMANCE.mdActive Frontier (+6 commits): Application projection memoization.
56perf-read-proof01a10490PROGRAMMING-PERFORMANCE.mdLanded: Diagnosed & capped the 65 KB / 1,016-item per-read proof overhead.
57powerbox-picker01a10498SPK.mdLanded: Cross-grain capability powerbox request/fulfill flow.
58pred-index-btree01a10505PROGRAMMING-PERFORMANCE.mdLanded: Secondary relation indexing for Pred queries (Law 17).
59proj-uwueave-v201a10512INTEGRATION.mdPreserved in minidregg-wt: codex-uwueave-projection-v2 worktree.
60room-template-workroom01a10520EXPERIENCE-ROOMS.mdLanded: Bundled workroom template (channels + docs + board + Hermes).
61search-fts-cell01a10528DOCUMENTS.mdLanded: Incremental trigram/token index over authorized cells.
62seat-board-coord01a10535ORCHESTRATOR.mdLanded: board/seats/*.md non-overlapping file-ownership protocol.
63selvage-fri-gap01a10542REVIEW-STRANDS.mdLanded: Discharged correlated agreement / proximity step in Selvage/.
64sync-merkle-pull01a10550SERVICE.mdLanded: Incremental follower sync over Merkle-diff ranges (Law 13).
65un-deputy01a10558RESIDENT.mdActive Frontier (+9 commits): Confused-deputy elimination across grain calls.
66web-shell-pwa01a10565EXPERIENCE-ROOMS.mdLanded: Offline-capable web client shell with local receipt verification.
67witness-gen-derived01a10572REVIEW-STRANDS.mdLanded: Derived witness table emission from Compiler/TraceTable.lean.

3.3 Commits e4a5b6c, 6c45f2e, 05dfcf9, and e1a3d25: Prospective Privacy, Compute-Funding v9, and the 216-Second Workroom Trace

Three landmark commits at the top of claudesplosion sealed the contracts between minidregg's kernel and its live deployment:

  1. **Commit e4a5b6c ("Preserve native prospective document protection receiving evidence", 2026-10-02)**:
  2. Documented in /Users/ember/dev/claudesplosion/planning/docuverse-convergence-2026-10-02.md, this commit unified two competing observation protocols into a single canonical tagged union (OBSERVE-CHALLENGE/v9 and OBSERVE-SIGNED/v9) and enforced prospective document encryption in native/spk-host/src/content_privacy.rs and native/spk-host/src/protected_object/:
  3. Every document mutation action (createDocument, editElement, createAtom, editAtom) is classified at the ingress boundary by content_privacy.rs.
  4. If a protected room's document action arrives with cleartext body atoms instead of an epoch-and-roster-bound ciphertext envelope (protected_object::SealedEnvelopeV1), admission refuses immediately before any bytes reach the journal.
  5. **Commit 6c45f2e ("Checkpoint historical carry, credential and compute-funding contracts", 2026-10-03)**:
  6. In /Users/ember/dev/claudesplosion/planning/final/transaction-v9-compute-contract-2026-10-02.md, Ember froze DREGG/RESOURCE/TRANSACTION v9 tag 6:
text
     Tag 6: computeFunding(
       asset                : AssetId,
       credits              : Nat,
       expectedPayerBalance : Nat,
       expectedBookRoot     : Digest
     )

This constructor atomically debits credits of asset from the caller's Book balance (requiring exact CAS match on expectedPayerBalance and expectedBookRoot) and mints a turn-scoped compute allowance consumed by Hermes model inference or .spk grain execution.

  • In /Users/ember/dev/claudesplosion/planning/final/native-carry-construction-2026-10-02.md, Ember froze the Public5688 IMAGE/v1 carry lift, preserving 5,688 historical public cells and their provenance digests across epoch re-genesis.
  • **Commits 05dfcf9 & e1a3d25 ("Add Codex handoff entrypoint for Mini", 2026-10-03)**:
  • In /Users/ember/dev/claudesplosion/planning/handoff-2026-10-03/START-CODEX.md, Ember handed minidregg (e851920e) to Codex with a sealed 24-role runtime capsule family-90562d76 (Host binary SHA-256 f6500290… compiled from Lean e8582196, exposing 168 active and 2 reserved host operations).
  • START-CODEX.md also captured the exact performance root-cause trace of why dregg room new --template workroom took 216 seconds on cold start:
  • Creating a single workroom template triggered 1,080 sequential Host IPC exchanges;
  • Spawned 94 fresh subprocess invocations of the standalone verifier binary; and
  • Generated 65 KB (1,016-item) inclusion proofs on every single cell read because the store serialized the full unpruned sparse-map spine per read.
  • This exact forensic trace directly drove the October perf-client, wave-host-light, iter-speed, and nanodregg CHARTER.md ("THE BAR: 1000 records, write p99 < 15ms, cold reopen < 200ms") engineering waves.

4. Excavating /Users/ember/dev/minidregg-lanes (273 Lanes) & /Users/ember/dev/minidregg-wt

To scale development across kernel semantics, Objective Bend compilation, the Selvage BaseFold proof system, Sandstorm .spk application qualification, and the Hermes autonomous resident without lock contention on a single checkout, Ember operated /Users/ember/dev/minidregg-lanes alongside /Users/ember/dev/minidregg-wt and the non-overlapping seat ownership protocol in /Users/ember/dev/minidregg/board/seats/.

4.1 Census & Domain Taxonomy of the 273 Entries in minidregg-lanes

An exhaustive git-object and working-tree census of /Users/ember/dev/minidregg-lanes against minidregg main (c4ecabe0f, October 31, 2026) establishes the exact partition of all 273 entries:

  • 140 Clean Merged Git Worktrees: Fully merged into main (c4ecabe0f) with 0 uncommitted or untracked files. These cover the completed waves of app-batch-1..12, app-market-b1..b2, app-market-b4, app-market-b6, fae-1..2, fae-4..7, fae-10..12, kn1..kn4, pd-auth, pd-book, pd-doc, pd-rooms, red-*, and snipe-*.
  • 67 Merged-but-Dirty Git Worktrees: Their committed HEAD is an ancestor of main (c4ecabe0f), but their working directories preserve uncommitted diffs, reproducer logs, or untracked Lean/Rust modules (dirty=1 to dirty=1019).
  • 38 Unmerged minidregg Git Worktrees (+1 to +24 commits ahead of main) (plus 2 foreign-repo worktree mounts with +0 minidregg commits): Active feature waves, November 2 consolidation branches (wave-apps-workspace, wave-assurance-detwin, wave-host-light, wave-prose-seed, perf-client), and deep qualification branches (hermes-live +24, oracle-v7-qualification +23, iter-speed +16, un-deputy +9, app-reds-3 +6, perf-apps +6).
  • 26 Non-Git Entries: 23 isolated Cargo/Lake build caches (target, target-demo-reds, target-integ, target-integ-pb, target-integ-spk, target-iter-speed, target-pb, target-probe, target-proofs, target-spk-fresh, fae-5-target, fae-6-target, no-fouc-base-target, no-fouc-target, pd-feedback-target, pd-grain-class-target, pd-host-rate-target, pd-invite-target, pd-private-compute-target, pd-web3-target, glm-spk-scout, hp-tools, rescue-2026-10-28) and 3 backup files (fae6-commit.txt, fae6-main.bak, fae6-rr.bak).

4.2 Complete Disposition Table of the 38 Unmerged minidregg Worktree Lanes

Every unmerged minidregg lane in /Users/ember/dev/minidregg-lanes, ordered by functional wave and commit divergence from main (c4ecabe0f):

#Lane WorktreeHEAD SHAAheadDirty FilesLast Commit Date & Subject
1wave-apps-workspace00d88b467+1402026-11-02: wave3-hermes-apps-tick: fix bend_answer @file staging, law_check...
2wave-host-light132b4e178+1402026-11-02: host/kernel: port lookupLight, FnEvidence, FnArchivePublisher to...
3wave-prose-seedf81134c8b+1402026-11-02: resource-client: fix SeatJson/ObjectiveBendDataWire wire compliance
4wave-assurance-detwin8c42ba8b5+1352026-11-02: assurance,verify: add op 151 receipt-opening and certifying-solver
5perf-client17a7c0811+1202026-11-02: world-overhaul: infer DrexBatch clear market, wire POST /act/surface
6hermes-live048b6b89d+2402026-10-27: J14: after dismissal the tail names Hermes by its short subject
7oracle-v7-qualificationb32f15604+2302026-10-28: Checkpoint Oracle qualification after approved host deletion
8iter-speede8a27db8d+16632026-10-27: iter-speed: state file, successor steps
9un-deputy4f99aa551+91092026-10-28: Deputy state: resume note after stand-down (stale re-wake corrected)
10app-reds-30979c697d+602026-10-31: consent worker: a connection past the served bound waits for a slot
11perf-apps76e22c9d2+602026-10-31: wip: doc batch 5
12codex-review-authority1f874f92c+322026-10-27: Clarify public-reader exposure and preserve review probes
13qualify-doc-contentiona0b45fe8e+302026-10-28: docs: COMMUTING-APPEND-STATE.md: the statements elaborate on dregg-ref
14un-ns-owner6216374ab+3322026-10-28: ROUTE-LATENCY-STATE: rewritten whole as present state; the owed...
15app-market-b533ee01407+2282026-10-29: checkpoint: save uncommitted work (root, 2026-10-30, before traversal)
16he-enrol32de8a4d4+202026-10-31: wip plain dispatch light
17hot-path-compact-62368f66d6cc2+202026-10-27: open: compaction reaches every cell's own store, not only the World
18oracle-codex-recovery0903ae21f+202026-10-27: Recover the five Oracle files' edits discarded by the 13:00 UTC reset
19untracked-2026-10-27d9771aa8b+202026-10-27: Triage table for the untracked files in the shared worktree
20app-market-b3d29428fa7+1352026-10-28: B3-STATE.md: the share-window statement is parked by the demo freeze
21app-reds646c7fd42+1112026-10-31: typed applications: keep the text of an unclassified controller
22audit-platform602f68f66+102026-10-28: Thirty-first pass: share-issue plan window
23codex-review-semantics293dcf241+152026-10-27: Review ordinary selection, scalar display and historical content
24cx-regen5f0f5cb1d+112026-10-29: checkpoint: save uncommitted work (root, 2026-10-30, before traversal)
25hot-path-world-checkpoint7b2e8fecd+102026-10-29: checkpoint: save uncommitted work (root, 2026-10-30, before traversal)
26hp-effects567d79eb8+102026-10-29: checkpoint: save uncommitted work (root, 2026-10-30, before traversal)
27infra-composed-root9f9ade67f+102026-10-28: composed residents: the two root hand-offs, performed only beside...
28pd-grain-class4ef18de80+142026-10-29: checkpoint: save uncommitted work (root, 2026-10-30, before traversal)
29perf-rust4142e1902+102026-10-28: perf-rust: handoff state for the in-session tool operations
30proofs-rerecord2fd8e091d3+102026-10-30: rerecord.sh: a red fixture is stale data; cover the doc-write rule
31qualify-sharefix1ed5a180f+102026-10-29: checkpoint: save uncommitted work (root, 2026-10-30, before traversal)
32scout60a9a5451+102026-10-30: scout: retract J14 red (stale box.env test provider in the scout)
33statement-fixes72722e54f+102026-10-29: checkpoint: save uncommitted work (root, 2026-10-30, before traversal)
34un-bend-seede65e5dfbd+112026-10-28: Tariff 6/2 re-run note: what passed and what the demo freeze stopped
35un-ifaddrs-shimf2495cad8+102026-10-28: docs: state of the getifaddrs shim lane (the IPython write row is...
36un-public-read630e50c21+1112026-10-28: docs: public-read lane state (403 fix, cold-start 504 findings)
37un-share-windowd08d8c1b0+112026-10-28: share-issue window statement: elaborated on dregg-ref, Accepted
38un-story-installcf5602792+102026-10-28: Park note for lane/un-story-install: runs owed and the unexplained...

4.3 Complete Table of the 67 Merged-but-Dirty minidregg-lanes Worktrees

The following 67 worktrees have their HEAD commit merged into main (c4ecabe0f), but hold uncommitted modifications or untracked artifacts that must be triaged before worktree pruning:

Domain ClusterWorktree Lanes (HEAD SHA / Dirty File Count / Preserved WIP Focus)
Sandstorm App Market & Grain Runtime (7 lanes)app-market-b7 (886a1ced5, dirty=23, WBO hook diagnostics); app-market-b8 (20f2ea776, dirty=51, failed-app resident evidence capture); app-market-fixups (299c0cb98, dirty=7, HakyllCMS system locale fix); app-reds-2 (40efdcff9, dirty=11, non-zero u64 volume resource IDs); app-writes-2 (8a3682c3e, dirty=2, Hacker Slides slides.md publish); un-spk-strict-sig (9f37f7022, dirty=4, strict SPK package stream signature verification); un-worker-concurrency (dd7de9faf, dirty=4, concurrent member serving during grain install).
Oracle Appliance & Enrollment (10 lanes)cx-oracle (0b0cbbf2f, dirty=75); oracle-cold-replay (72d8c569a, dirty=17, historical replay comparison at Store epoch boundary); oracle-continuation (9d66bde72, dirty=17, acceptance world & shared-data); oracle-convergence (7be0767ba, dirty=1, Oracle appliance structural review); oracle-enrollment (9d66bde72, dirty=11); oracle-enrollment-review (30bd2387f, dirty=3, retained pre-tool refusal recognition); oracle-grant-factory (7470f534a, dirty=1, local Host selection); oracle-hosted-continuation (5dafe0f3d, dirty=62); oracle-runtime-v6 (5dafe0f3d, dirty=6); oracle-served-inspection (365439e47, dirty=11, canonical room request identity join); oracle-tenant-bridge (0c63478d5, dirty=15).
Private Backend & Threshold Custody (un-pb-* + pd-*, 11 lanes)un-pb-codec (470f41e7a, dirty=4, gate row R8 Lean codec agreement); un-pb-exec (642113406, dirty=2); un-pb-helpers (ee908664b, dirty=26, P6 lost-reply recovery on reserve/burn); un-pb-seal (e8fd70191, dirty=12, custody epoch strictly from Host); un-pb-twins (134535c23, dirty=18, blocking accept + wake socket in custody); un-pb-wal (68d9f28ea, dirty=14, King composite & ACSS-Id dealer WAL commit); pd-feedback (776756637, dirty=5); pd-host-rate (9f9bac798, dirty=8, Host.Main ratedClaims allocation gate); pd-invite (567e8db7b, dirty=6); pd-private-compute (a5b6cb6b6, dirty=10); pd-web3 (e4f3904cf, dirty=7).
Kernel, Effects, Pred & Assurance (un-*, 18 lanes)un-assurance (e6eac9957, dirty=17, Kernel/StoryLaw.lean table-to-law generator); un-bend (91f8a83ce, dirty=20, poll(2) lose-reply wait); un-effects (2acd8acf0, dirty=50, world-activity-admit native scenario commons); un-effects-rows (20f28ac31, dirty=4); un-fixtures (a6a321033, dirty=20); un-japps-sig (8ef4c65a9, dirty=1); un-moves (755e05332, dirty=22, KERNEL-FAMILIES DreggNet lease cluster); un-native-rows (6fae5fa4e, dirty=8); un-nhr-tag (753271d7c, dirty=2, NativeHostReplay.validateLoaded_headTag); un-pred-census (44479e30f, dirty=27, ReceivingVerifier: Vouchers.answered); un-seats (ccb34bc5e, dirty=1); un-shared-names (f0894a415, dirty=1); un-stream-continuity (f0df4bc24, dirty=1); un-strict-verify (1158dd071, dirty=5); un-tale-welcome (bc6c837ba, dirty=5); un-verify (9f5a7cf40, dirty=6, Generic Simplex gate); un-web-front (0da049992, dirty=1); un-world (ae6bf1a1a, dirty=16, effect Plan keystone).
Store Retention, Build Loop, Review & Web (20 lanes)retained (c4ecabe0f, dirty=20, store-retained persistent map cell state at main HEAD); exorcism (0b728dc32, dirty=2, board/seats/read-path.md); iter-speed-land (1d4f01724, dirty=1, codec census private import seam); build-loop-tools (85a8f4c8e, dirty=3); integration-r2 (5c177d3dc, dirty=1); cx-app-interface (842d8fb75, dirty=10, umbrellas stage); cx-store-sweep (0b0cbbf2f, dirty=1019, bulk store sweep artifacts); codex-cut13-review (b2c7dcb5c, dirty=1); codex-review-provider (8f6a6524a, dirty=15); fae-3 (e4b02fd9f, dirty=2); fae-8 (d13cc2f26, dirty=1); fae-9 (a505ba977, dirty=4); glm-a11y (12d06714c, dirty=5); glm-lean-scout (91cdd6d49, dirty=1); glm-native-scout (568205cd6, dirty=24); red-ktpl-template-links (4b2e5540c, dirty=1); red-mini-web-devflow (4be7afc38, dirty=1); snipe-redteam (4f3d2506b, dirty=1); snipe-welcome-hermes (31bb7201f, dirty=1); web-parity (0a4aefd49, dirty=1).

4.4 /Users/ember/dev/minidregg-wt (3 Git Worktrees + seats/)

Alongside minidregg-lanes, /Users/ember/dev/minidregg-wt holds 3 specialized cross-repo integration worktrees and the seats/ coordination directory:

Worktree PathHEAD SHA & DateWorking Tree StateArchitectural Purpose
minidregg-wt/codex-uwueave-projection-v2f149cb4cb (2026-08-11)Clean (dirty=0)Implements Compiler/UwueavePreoProjectionV2.lean and emits prover/generated/uwueave_preo_projection_v2.rs (CROSS_REPO_CONTRACTS.md §1).
minidregg-wt/fn-e2-materialized-cache056846cef (2026-09-23)Clean (dirty=0)Materialized projection cache for fn consumer queries over World cells.
minidregg-wt/fn-evidence183cd3769 (2026-09-24)dirty=4 (M Kernel.lean, M Kernel/FnConsumerFrontierProofs.lean, M Kernel/FnConsumerFrontierScenario.lean, ?? Kernel/FnGatewayPolicy.lean)Formal gateway policy and consumer frontier proofs (FnGatewayPolicy.lean), actively being ported into main via minidregg-lanes/wave-host-light (132b4e178).

5. /Users/ember/dev/nanodregg (871 Commits), /Users/ember/dev/nanodregg-scout (27 Scouts), and /Users/ember/dev/nanodregg-wt

On October 16, 2026, Ember initiated /Users/ember/dev/nanodregg (871 commits through October 27, 2026, HEAD 80c622a)—a radical, spec-first, single-writer distillation of minidregg governed by a 12-law charter, a Critique/Meta-Critique ledger, and an automated size and theorem ratchet.

5.1 The 12 Charter Laws (CHARTER.md) and the Monotonic Size Ratchet (SIZE.md)

/Users/ember/dev/nanodregg/CHARTER.md establishes 12 inviolable laws:

  1. Law 1 (One of Everything): One representation per concept, one wire format (Codec/), one hash/MAC family (cSHAKE256 / KMAC256), one signature scheme (RFC 8032 Ed25519), one store format (seg.log + sealed bases), one language (Objective Bend in Lang/).
  2. Law 2 (Lean Decides; Rust Moves Bytes): Every policy, framing bound, retry budget, and admission rule lives in Lean (Kernel/, Lang/, Protocol/, Host/, Hosting/) or is emitted into Rust (Emit/Rust.lean → client/src/table.rs). The Rust binaries (nd at 99 lines, nd-supervise at 164 lines) contain zero policy branches.
  3. Law 3 (Bytes End-to-End): Proofs do not stop at abstract ASTs; Codec/Proofs/ proves round-trip invertibility (decode (encode x) = .ok x) down to raw ByteArray streams.
  4. Law 4 (TigerStyle Total Fuel-Bounded Loops): Zero partial definitions in deployed Lean modules (0 across all 32,077 lines); every loop is bounded by an explicit structural fuel or protocol Bounds constant.
  5. Law 5 (Performance Designed — THE BAR): Every full judge runs scripts/bar.sh driving two confined members and one confined writer through real sockets over 1,000 records, enforcing write latency and cold reopen budgets.
  6. Law 6 (Named Theorems, No #guard, Tree-Wide Axiom Census): Every claim is a named theorem inspected by gates/ across the entire dependency cone.
  7. Law 7 (Refusals by Name): Every error across all layers is an explicit constructor of a sum type (86 refusal constructors across Kernel.Reject, Lang, Protocol.Fault, Host.Exit, and Hosting), never a string.
  8. Law 8 (Split Proofs by Proofs/ File): Implementation modules (Kernel/World.lean, Protocol/Rows.lean) stay compact and readable; proofs live in */Proofs/*.lean (287 proof modules out of 529 total Lean files).
  9. Law 9 (Every Commit Builds via CAS scripts/commit): Commits land only through a compare-and-swap build judge that compiles every target and runs the theorem/axiom census.
  10. Law 10 (Decisions + Critique + Meta-Critique): Every architectural choice in DECISIONS.md and decisions/*.md (HOST-1..7, CLIENT-1..39, KERNEL-1..66, PROTOCOL-1..72, HOSTING-1..48) includes an adversarial Critique and a Meta-Critique before code is written.
  11. Law 11 (No Twins of minidregg).
  12. Law 12 (The Size Ratchet — SIZE.md): /Users/ember/dev/nanodregg/SIZE.md at 80c622a records the exact audited footprint:
Metric in nanodregg/SIZE.md (80c622a)Exact CountNotes
Total Lean Modules52995 in host executable closure; 287 in Proofs/
Deployed Lean Lines32,077Across Codec/, Kernel/, Lang/, Protocol/, Host/, Hosting/, Emit/
Named Lean Theorems6,703Audited by gates/tcb.tsv and axiom census
partial Definitions / sorry / Unapproved Axioms0 / 0 / 0Strictly zero across the entire repository
@[extern] C Bindings & C Lines12 / 418 linesMinimal OS socket/fsync/clock primitives
Rust Lines (nd CLI / nd-supervise)99 / 164 linesDriven by Emit/Rust.lean generated table.rs
Protocol Operations / Receive.Body Kinds / CLI Verbs7 / 13 / 20Compact, closed surface
Total Named Refusal Constructors / Conformance Rows86 / 52Tested positive and negative (with fault plants)

5.2 Inside Kernel/World.lean, Protocol/Rows.lean, and Spec/README.md

In /Users/ember/dev/nanodregg/Kernel/World.lean, the world state is represented by three persistent binary Merkle tries (Cells, Journal, WindowSlot) indexed over explicit bit-depth keys, and admission returns a closed 39-constructor Reject sum type:

lean
-- /Users/ember/dev/nanodregg/Kernel/World.lean:24-142
structure Env where
  digest   : ByteArray → Digest
  mac      : Key → ByteArray → Digest
  verify   : PubKey → ByteArray → Sig → Bool
  checkLaw : LawPkg → Value → Value → CallCtx → Except LawRefusal Unit

structure Slot where
  owner     : PubKey
  lawDigest : Digest
  valDigest : Digest
  value     : ByteArray
  nonce     : UInt64
  allowance : UInt64
  deriving DecidableEq, Repr

/-- Persistent binary Merkle tries for cells, journal entries, and replay window slots. -/
abbrev Cells      := Trie 64 Slot
abbrev Journal    := Trie 64 Entry
abbrev WindowSlot := Trie 16 ReplayRecord

/-- Closed 39-constructor kernel refusal sum type (Law 7: Refusals by name). -/
inductive Reject where
  | badSig | staleNonce | retiredCell | unknownCell | cellAlreadyExists
  | quotaExceeded | allowanceUnderflow | lawRejected (r : LawRefusal)
  | pinMismatch | rowPinMismatch | rangePinMismatch | windowExpired
  | duplicateCall | invalidGenesis | ... -- 39 named constructors total
  deriving DecidableEq, Repr

In /Users/ember/dev/nanodregg/Protocol/Rows.lean, the 52 conformance rows are first-class Lean data structures (Row) that both drive the live socket test runner and serve as the subjects of formal proofs in Protocol/Proofs/:

lean
-- /Users/ember/dev/nanodregg/Protocol/Rows.lean:18-76
inductive Expect where
  | admitted (height : Nat)
  | refused  (reason : RefusalName)
  | followed (height : Nat) (root : Digest)
  deriving DecidableEq, Repr

inductive Fault where
  | none
  | truncateTail (bytes : Nat)
  | flipBit (offset : Nat)
  | dropFrame (index : Nat)
  | withholdPrefix (upTo : Nat)
  | staleReRun
  deriving DecidableEq, Repr

Most remarkably, /Users/ember/dev/nanodregg/Spec/README.md defines four top-level end-to-end system security properties—OneHistory, Integrity, Consent, and Shown—and requires two paired theorems for every property:

  1. top_* (Soundness of deployed): Proves that the real deployed system satisfies the claim (top_oneHistory_member, top_oneHistory_pair, top_integrity_value, top_integrity_law, top_consent_okCommitted, top_consent_final, top_consent_cancelOk, top_consent_staleFinal, top_shown_onlyFollowers).
  2. floor_* (Non-Vacuity / Refutation of trivial): Proves that a lazy trivial implementation fails the exact same statement (floor_oneHistory_member, floor_integrity_law, floor_consent_okCommitted, etc.), mathematically guaranteeing that no specification predicate in Spec/ is vacuously true!

5.3 Complete Excavation of the 27 Architectural Scouts (/Users/ember/dev/nanodregg-scout)

Before landing features into nanodregg, Ember dispatched 27 isolated scout prototypes in /Users/ember/dev/nanodregg-scout (C1..C2, D1..D3, G1..G2, H1..H5, L1..L4, O1..O5, P1..P3, T1..T2), each containing a standalone SCOUT.md analysis and executable Lean/Rust/C proof-of-concept:

ScoutFiles in /Users/ember/dev/nanodregg-scout/<ID>/Architectural Question & Prototype Built
C1Thin.lean, Demo.lean, SCOUT.mdThin consent via sealed base (HOST-15): Proved that a thin member adopting one verified whole-World base and folding only the log tail shares the single Host.Follow code path (PROTOCOL-14 option A).
C2Blind.lean, All.lean, SCOUT.mdPer-cell hiding: Designed blinded cell slots in the log where followers verify Merkle root transitions without holding the per-cell blinding key.
D1ARCHITECTURE.md, Boundaries.lean, SCOUT.mdCode-derived architecture map: Automated import-graph and boundary signature extractor across all nanodregg layers.
D2MATRIX.tsv, THEOREMS.tsv, REFUSALS.tsv, GAPS.txt, SCOUT.mdThe Spec Matrix: Cross-tabulated every row in Protocol/Rows.lean against THE BAR, covering theorems, fault plants, and WASTE rows.
D3TUTORIAL.md, Tour.lean, Explain.lean, SCOUT.mdHermes resident tutorial: Zero-to-installed-law tour using only nd, host follow, and Lang/Read, with every refusal named and explained.
G1Laws.lean, Pkgs.lean, E2E.lean, SCOUT.mdStandard library of laws in Objective Bend: Built checked core terms, package bytes, and teeth for ownerOnly, bounded, quorum-of-n, rate-limit, and allowlist.
G2Cross.lean, SCOUT.mdActivities, perform, and cross-object calls (LANG-13): Compared synchronous in-language perform call vs. multi-turn message passing and powerbox exercise.
H1Grain.lean, GrainPlan.lean, GrainPkg.lean, SCOUT.mdGrain lifecycle as Lean-decided policy: Modeled install, start, stop, and volume attachment as Receive call kinds with cross-store volume refusal by name.
H2lean/, rust/, policy.bin, SCOUT.mdProvider-key broker (KEY-CUSTODY option A): Built a Lean decider + Rust byte-mover broker so model-provider API keys never enter a hosted resident's address space.
H3attach/, fakehost.py, drive.py, SCOUT.mdnd-attach over SSH (CLIENT-12 option B): Multiplexed conn-tagged streams over ssh -T with restrict,command="nd-attach SOCK" per friend key.
H4H4Resident.lean, H4Theorem.lean, SCOUT.mdMetaprogrammable Hermes resident loop: Hosted resident authors an Objective Bend law as s-expressions, checks it offline via checkWith, and submits it for admission.
H5Repl.lean, session.ob, SCOUT.mdResident check / REPL CLI for Objective Bend: Interactive s-expression checker, package-id calculator, and fuel-bounded evaluator.
L1Ledger.lean, LedgerProofs.lean, SCOUT.mdLedger (accounts, assets, transfer): Implemented double-entry transfers as Kernel.Receive call kinds with a balance conservation theorem modeled on admit_charge.
L2Ledger.lean, T1.lean, SCOUT.mdWells, issuance caps, and credit floors (pug issue #1): Proved outstanding_le_sum_floors over setFloor and sweep operations.
L3PayBook.lean, PayDsl.lean, SCOUT.mdPay journeys (JPAY1–6): Restated purse, provider routes, fees, and settlement as nanodregg conformance rows and receivers without legacy shapes.
L4Escrow.lean, NullOnce.lean, EJourney.lean, SCOUT.mdBounty / escrow receiver + law: Proved that escrowed funds are never both released on law verdict and refunded on expiry.
O1Deploy.lean, SCOUT.mdDeployment units & workhorse runbook: Generated systemd/launchd service units for host serve, host follow, nd-supervise, and the key broker.
O2Carry.lean, Perf.lean, SCOUT.mdWorld-carry and epoch re-genesis: Exported a Store fold and re-imported it under a new epoch with an equivalence proof against the exported state.
O3Backup.lean, SCOUT.mdStore backup & restore verification: Verified that restore checks open + MAC + chain and refuses truncated or spliced backups (seg.log + bases) by name.
O4Obs.lean, ObsRun.lean, SCOUT.mdTheorem-adjacent observability: Implemented standing lines, WASTE and BAR counters, and per-request loop timing with zero logging framework dependencies.
O5linux-standing.sh, sandbox-linux.patch, SCOUT.mdFresh Linux box standing verdict + bwrap audit: Tested bc0460d on Ubuntu 24.04.5 (burst5-a, 32 cores) and produced sandbox-linux.patch fixing live bwrap execution.
P1HostScout.lean, probe.py, lag.py, SCOUT.mdTHE BAR end-to-end over sockets: Profiled 1,000 records across host serve, host follow, and two nd clients on M2 Max, isolating write p99 and cold reopen bottlenecks.
P2MemoProgram.lean, LoadMemo.lean, SCOUT.mdThe load memo (LANG-16): Eliminated $O(\text{closure bytes})$ re-decoding in Lang.load on every judgeOne and runMethod call via content-addressed memoization.
P3sync.c, bench/, SCOUT.mdBase MAC at THE BAR scale: Benchmarked compiled Codec.Hash.mac (KMAC256) over 290 KB and 29 MB World encodings against fsync latency (3.9–6.4 ms) to set base-sealing cadence.
T1Multi.lean, Plants.lean, SCOUT.mdMulti-Store routing & cross-store refusal (HOST-16): Proved that a frame addressed to another Store is refused by name (storeMismatch) at the supervisor router.
T2Child.lean, src/, tests/, SCOUT.mdRemote socket mode of nd-supervise: Audited mux framing, connection tags, CONNECTIONS_MAX, and idleMsMax over forwarded SSH sockets, identifying 3 60-line fixes later landed in CLIENT-38/39.

5.4 Complete Audit of the 9 Worktrees in /Users/ember/dev/nanodregg-wt

/Users/ember/dev/nanodregg-wt contains 9 git worktrees—five late-October repair lanes (fixhosting, fixkernel, fixproto, fixrust, fixrust4) addressing the October 24 Collation findings (COLLATION-20261024), and four early layer worktrees (host, perf, protocol, root) from October 16–17:

WorktreeHEAD SHAWorking Tree StatusKey Commits & Uncommitted Changes
nanodregg-wt/fixhosting37772b9Dirty (10 M, 2 ??): Hosting/Proofs/{BellConform,Lifetime,ResidentRun,Socket,SocketStrong,FollowAnswer}.lean, Hosting/Resident/{Bell,Run,Trace}.lean, decisions/hosting.md, gates/tcb.tsvHOSTING-48 (37772b9): Proves no_refusal_repeats_forever over Ticks (bounding resident retry loops by retriesMax and ending with retryExhausted). Working tree holds uncommitted Hosting/Proofs/FollowAnswer.lean.
nanodregg-wt/fixkernel0a299c8Dirty (2 M): Lang/Journey.lean, decisions/kernel.mdAt PROTOCOL-70/71/72 (0a299c8), holding uncommitted edits to Lang/Journey.lean and decisions/kernel.md.
nanodregg-wt/fixproto0a299c8Dirty (7 M): Protocol/Command/Machine.lean, Protocol/Proofs/{Behind,Command}.lean, Protocol/Rows{,.lean,/Judge.lean,/Run.lean}, decisions/protocol.mdHolds in-flight protocol proof refinements on Command/Machine.lean and Behind.lean above 0a299c8.
nanodregg-wt/fixrustb869bb91 ahead + Dirty (1 ??): ?? decisions/client.md.card2b869bb9 ("WIP cards 2+3 (FIX-RUST, local only)") above 1ab410e (CLIENT-38: die_with_parent via PR_SET_PDEATHSIG / getppid watch, process-group kill on row timeout).
nanodregg-wt/fixrust4234cb4d1 ahead, Clean (dirty=0)234cb4d ("WIP card 4 (FIX-RUST-2, local only)") above 2d5056e (CLIENT-39: one frame::Deadline per reply, deciderLost by name past Broker.Bounds.answerMsMax, atomic fsx::create_once).
nanodregg-wt/host28de15eDirty (13 M, 1 ??): Host/{Cli,Exit,Follow,Loop,Serve,Store,Wire}.lean, ?? Host/Proofs/Loop.leanEarly Oct 17 worktree (D5 exit table as data & D6 host check STORE --anchor); superseded on main but preserves Host/Proofs/Loop.lean draft.
nanodregg-wt/perfb616a8dClean (dirty=0)Early Oct 17 worktree (PERF-1: scripts/bar.sh socket benchmark); already merged into main.
nanodregg-wt/protocola02b35cDirty (10 M, 2 ??): ?? Protocol/{Attempt,Proofs/Attempt}.leanEarly Oct 17 worktree (PROTOCOL-14 thin consent); superseded by Protocol/Command/Machine.lean on main.
nanodregg-wt/root09ff240Dirty (1 M): M CHARTER.mdEarly Oct 17 worktree (ROOT-HBOX landing recipe); already merged into main.

6. /Users/ember/dev/delvetalk & /Users/ember/dev/delvetalk2 (2,583 Commits) & /Users/ember/dev/delvetalk2-lanes (66 Lanes)

6.1 From delvetalk v1 to delvetalk2 (delve.town, Replay Edition 20)

/Users/ember/dev/delvetalk (21 commits, HEAD 9f70515) began as an early prototype of a Smalltalk/MUD-inspired object-capability environment backed by Lean. In /Users/ember/dev/delvetalk2 (2,583 commits, HEAD fb2179968 on October 15, 2026), Ember built the full production system powering delve.town (https://gsb.fg-goose.online, replay edition 20) and its open feed-only sister world (https://obend.fg-goose.online).

As documented in /Users/ember/dev/delvetalk2/README.md, a town in delvetalk2 is an append-only DAG-CBOR journal of object turns. Every resident, room, bell, commonplace book, automaton, and rule is an Object with four components:

  1. Identity (id): A human-meaningful path such as garden, bell/1, avatar/ember, or kind/bell.
  2. Content-Addressed Objective Bend Closure (pin): Built from world/lib/, the 9 core object archetypes in world/objects/ (Record.obend, Thing.obend, Kind.obend, Avatar.obend, Library.obend, Mending.obend, Policy.obend, Tide.obend, Workshop.obend), and the 18 domain bodies in world/bodies/ (automatafl.obend, bell.obend, board.obend, clock.obend, commonplace.obend, counter.obend, deal.obend, door.obend, fountain.obend, garden.obend, guestbook.obend, lantern.obend, letter.obend, loom.obend, mirror.obend, oracle.obend, shelf.obend, weathervane.obend), composed via Faré-style prototype inheritance (fix / mix) and Spween declarative scene graphs (world/station.spween).
  3. Canonical DAG-CBOR State (stateCid): Conforming to the object's compiled Objective Bend state type and relational uniqueness keys.
  4. Modal Clause Text (law / rule / want): Parsed and denoted by a single unified grammar in spec/bend/Compiler/ObjectiveBendLaw.lean.

6.2 The Unified Modal Clause Language (ObjectiveBendLaw.lean)

In /Users/ember/dev/delvetalk2/spec/bend/Compiler/ObjectiveBendLaw.lean, every object carries line-oriented clauses in three modalities:

  • law NAME "reading": EXPR — Refuses a write at commit time unless $\varphi$ holds ([write]φ).
  • rule NAME "reading": EXPR then SPELL — Fires at commit time under the rule owner's authority when $\varphi$ holds ([write]φ then σ), casting SPELL (delvetalk ... with interpolated {TERM} holes).
  • want NAME "reading": EXPR asks SPELL — Poses a standing desire evaluated at read time against the world clock (φ standing), surfacing SPELL on the object's card and the town's daily chronicle as an action wanting a hand.
lean
-- /Users/ember/dev/delvetalk2/spec/bend/Compiler/ObjectiveBendLaw.lean:89-190
inductive LawFact where
  | subject | caller | height | turn | pin | kind | method | verb | clock | object | laws | refusedBy
  deriving DecidableEq, Repr, Inhabited

inductive LawSide where
  | new | old | cited (object : String)
  deriving DecidableEq, Repr, Inhabited

mutual
inductive LawTerm where
  | field (side : LawSide) (name : String)
  | fact (fact : LawFact)
  | count (side : LawSide) (field : String) (filter : LawFilter)
  | int (value : Int)
  | text (text : String)
  | sum (left right : LawTerm)
  | mod (term : LawTerm) (modulus : Nat)

inductive LawFilter where
  | all
  | eq (column : String) (value : LawTerm)
  | and (left right : LawFilter)
end

inductive LawRule where
  | monotone | writeOnce | appendOnly | unchanged | insertOnly
  deriving DecidableEq, Repr, Inhabited

inductive LawExpr where
  | eq (left right : LawTerm) | ne (left right : LawTerm)
  | lt (left right : LawTerm) | le (left right : LawTerm)
  | gt (left right : LawTerm) | ge (left right : LawTerm)
  | member (term : LawTerm) (place : LawPlace)
  | rule (rule : LawRule) (field : String)
  | changed (field column : String) (who : LawTerm)
  | changedTerm (term : LawTerm)
  | under (object : String)
  | wants (object name : String)
  | not (body : LawExpr) | and (left right : LawExpr)
  | or (left right : LawExpr) | implies (premise conclusion : LawExpr)
  deriving DecidableEq, Repr, Inhabited

6.3 Stroke-Verified LLM Enchantment (Host/Enchant.lean) and Cross-Edition World Carry (Host/Carry.lean)

Two signature subsystems in spec/Delvetalk/Host/ solve the twin problems of hallucination-free LLM narrative rendering and zero-loss live world upgrades:

  1. Stroke-Checked Enchantment (/Users/ember/dev/delvetalk2/spec/Delvetalk/Host/Enchant.lean):
  2. Every act on a thing leaves structured, host-assembled Strokes (Line, Mark, Quantity, Verb, Rule, Want, Door, tone, fields) and span-bounded Telling records (away since the reader's last departure or before covering the place's prior history).
  3. When the world's model paints a prose card (maxPaint = 400, maxDay = 600, maxAway = 300, maxBefore = 500), Enchant.verify checks the generated prose against Strokes: every number must match an authorized Quantity beside its unit, every quoted phrase or handle must come from Strokes, and the exact factual card always stands beneath the paint.
lean
   -- /Users/ember/dev/delvetalk2/spec/Delvetalk/Host/Enchant.lean:87-120
   structure Telling where
     form    : String
     since   : Nat
     upto    : Nat
     exact   : List String := []
     history : String := ""

   structure Strokes where
     object  : String
     version : Nat
     kind    : String := ""
     voice   : String := ""
     about   : String := ""
     head    : String := ""
     fields  : List (String × String) := []
     tone    : String := ""
     maker   : String := ""
  1. Cross-Edition World Carry (/Users/ember/dev/delvetalk2/spec/Delvetalk/Host/Carry.lean):
  2. When the delvetalk-obend binary advances to a new replay edition (e.g., edition 19 → edition 20), delvetalk-obend carry reads the snapshot taken at the old journal's head (readWorld), recompiles every object against the new world/lib and world/bodies (rebuilt), runs package migrate(old: State8) -> State functions when state schemas evolve, verifies that every suspended continuation checkpoint decodes under the new compiled packet (resumable), and preserves every historical receipt slug (Host/Slug.lean, e.g. nosap-hodaz) so old links resolve seamlessly as "entry $N$ of the world before the carry".
  1. Multi-World Serving & AT Protocol Federation (/Users/ember/dev/delvetalk2/spec/Delvetalk/Serve.lean):
  2. Serve.lean serves the main persistent world, on-demand scratch worlds (isolated ephemeral forks where residents or agents can rehearse spells and laws before casting them in town), and a complete AT Protocol PDS/Repository (did:web:gsb.fg-goose.online) with k256 repo signing, Merkle Search Trees (MST), CARv1 export, and 18 online.fg-goose.gsb.* lexicons.

6.4 Complete Audit of All 66 Entries in /Users/ember/dev/delvetalk2-lanes

An automated git audit of all 66 entries in /Users/ember/dev/delvetalk2-lanes against delvetalk2 main (fb2179968, October 15, 2026) reveals an exceptionally clean worktree farm:

  • 44 Clean Merged Git Worktrees: All 44 feature lanes are 100% merged into fb2179968 with dirty=0:
  • activity, atproto, author, author-build, automod, bar-audit, bodies, bugdesk, carry-rehearse, chronicle, claude-overhaul, codex-carry, codex-ui, compile- trim, cx-world, deal-table, deploy-next, docs-audit, doors, enchant, enchant-tellings, fare-mix, federation, feed-world, garden-page, harmonic, kind-stack, law-clauses, lexicons, mending, migrate-v20, objects, objects2, observer, onboarding, policy-gate, ratchet, receipt-slugs, relations, rewrite14, scratch-worlds, spween-scenes, tide-clock, workshop.
  • 1 Unmerged Git Worktree (delvetalk2-lanes/perf):
  • At commit 5739d22ec6f0bc26e4ac20ad49f0b359e2300032 (+2 commits ahead of its October 9 base a08cb38f7, dirty=1 on spec/PerfBench.lean):
  • 996b498a8 ("scratch: perf-bench stage timer", adding spec/PerfBench.lean).
  • 5739d22ec ("The packet's knot holds only what the entry reaches"): Modifies spec/bend/Compiler/ObjectiveBendElaborate.lean (+71/-3) and spec/Delvetalk/Generics.lean (+20/-2) so elaborateM computes reachableKnot—the transitive closure of global declarations actually reached by an entry point (Output.knotRow) instead of packing all closure declarations into a single nested row. This lifts the 254-declaration closure ceiling to unbounded closure size (up to 254 reached definitions per entry), cuts Garden compile time from 0.82/0.96/0.83 s to 0.40/0.69/0.40 s, shrinks the compiled packet from 734 KB to 29 KB, and cuts test_await + test_turn_world + test_http from 174 s to 51 s.
  • 21 Non-Git Handoff Documents & Scratch Artifacts:
  • Key root handoffs: HANDOFF-root-2026-10-14.md (documenting edition 20 live deployment and post-overhaul invariants), author-build-handoff.md, cx-world-handoff.md, deploy-next-handoff.md, rewrite14-handoff.md, plus 16 benchmark logs and snapshot test fixtures.

7. /Users/ember/dev/drorb (618 Commits), /Users/ember/dev/drorb-swarm-salvage, /Users/ember/dev/DreggNet (520 Commits), and /Users/ember/dev/DreggCloud (88 Commits)

7.1 drorb (618 Commits): Verified Post-Quantum Edge Server & Tailscale Control/Data Plane

/Users/ember/dev/drorb (618 commits, HEAD ec01e68, October 6, 2026; detailed in GOAL.md) is a formally verified reverse proxy, TLS 1.3 / HTTP/2 / HTTP/3 / WebSocket edge terminator, and complete Tailscale (ts2021) control and data plane written in Lean 4, lowered through HOL4 Pancake (datacake), and driven by a thin Rust io_uring / kqueue ring I/O wrapper.

Key verified subsystems in /Users/ember/dev/drorb/:

  1. Verified Post-Quantum & Classical Cryptography:
  2. All 6 NIST PQ operations—ML-KEM-768 (keygen, encaps, decaps) and ML-DSA-65 (keygen, sign, verify)—plus FIPS 202 Keccak-f[1600] (rc_lanes_eq_exec proved by Lean kernel decide with zero axioms), X25519, ChaCha20-Poly1305, AES-GCM, and HKDF-SHA256/384.
  3. Verified HTTP/1.1, HTTP/2 (H2Ingress), HTTP/3, and Braid Calculus:
  4. Reactor/BraidCalculus.lean proves compositional non-interference (braid_gate, braid_transform) across 25+ middleware gates (forward-auth 401, connection-cap 503, stick-table 429, slowloris 408, conditional 304, redirect 308, method-filter 405, body-size 413, Host-allowlist 421, response-splitting and bare-LF smuggling rejection, and H2.Conn.feed_congr).
  5. Complete Tailscale (ts2021) Control & Data Plane:
  6. Implements the Noise_IK_25519_ChaChaPoly_BLAKE2s ts2021 control framing, DERP relay server/client over verified TLS, STUN binding responder, Disco ping/pong/call-me-maybe cross-NAT hole punching, IPAM (100.64.0.0/10), drorb-ctl CLI, and a verified HuJSON ACL compiler (servedMapFor_filter_is_acl_compile proving that the MapResponse packet filter sent to peers equals the compiled ACL policy).

7.2 /Users/ember/dev/drorb-swarm-salvage: The 7 Salvaged Lean 4 Modules & Braid-10 Patch

/Users/ember/dev/drorb-swarm-salvage preserves 7 standalone Lean 4 modules and swarm-tracked-changes.patch (50.6 KB) from the October swarm wave:

Salvaged FileLinesKey Verbatim Lean 4 Definitions & Theorems
QuicServer.lean1,786Full RFC 9000 / RFC 9001 QUIC v1 server state machine: initial salt 0x38762cf7f55934b34d179ae6a4c80cadccbb7f0a, packet number reconstruction, ACK range encoding (theorem ackRuns_exact), key-update phase bit rotation (theorem nextGenKeys_preserves_hp), and AEAD retry token validation (theorem checkRetryToken_sound).
Framing.lean342Verified HTTP/1.1 chunked framing, bare-LF rejection, and Server-Sent Events (SSE) framing (theorem wireBytes_eq_wireSpecLF, theorem sse_frame_parses_back).
DeployPlus.lean & DeployPlusDrive.lean418Composes braids 6 through 10 (braidedDeployment10: JWT 401, header-rewrite, SPA-fallback 200, request-validation 505, Date, framing-TE 400, Expect 417, auth-request 401, IP-filter 403, head-limit 431, not-impl 501, missing-Host 400, whitespace-name 400) and proves servePipelineOfMetered_braided10_off_eq.
CacheControl.lean214RFC 9111 Cache-Control directive parser/serializer and freshness lifetime calculator with round-trip and no-store dominance theorems.
ContentLocation.lean168RFC 9110 Content-Location URI-reference sanitizer and same-origin gate.
RetryAfter.lean182RFC 9110 §10.2.3 Retry-After delta-seconds parser/formatter with monotone clamping proofs.
swarm-tracked-changes.patch50.6 KBRe-points drorbServeMeteredBraided and drorbServeBraided in Dataplane.lean from braidedDeployment5 to braidedDeployment10 and wires QuicServer.lean into the ingress table.

7.3 /Users/ember/dev/DreggNet (520 Commits): Verified Pancake Reactor DSL & Execution Cloud

/Users/ember/dev/DreggNet (520 commits, HEAD 2c574b6; ARCHITECTURE.md) combines a 6-crate Rust execution cloud with a Lean 4 reactor compiler in Dsl/:

  1. Lean 4 Pancake Reactor Compiler (Dsl/Reactor.lean & Dsl/EmitPancake.lean):
  2. Models zero-copy io_uring / kqueue event loops as typed ReactorComponent state machines mapping RingEvent → RingSubmission, and compiles them to HOL4 Pancake ASTs (PExpr, PStmt, PFun, emitRegion):
lean
   -- /Users/ember/dev/DreggNet/Dsl/Reactor.lean:42-118
   /-- Copy-once buffer recycling law: every completed recv buffer slot is either
       recycled immediately into a send submission or returned to the free ring
       without double-free or leak. -/
   theorem reactor_prim_recycle
       (c : ReactorComponent) (s : RingState) (ev : RingEvent)
       (hwf : RingState.WF s) :
       let (s', sub) := c.step s ev
       BufferConserved s s' sub

   theorem reactor_prim_wf
       (c : ReactorComponent) (s : RingState) (ev : RingEvent)
       (hwf : RingState.WF s) :
       RingState.WF (c.step s ev).1
  1. The 6 Rust Cloud Crates:
  2. dreggnet-exec: Pure-Rust fuel-metered wasmi WebAssembly sandbox executing workload bundles.
  3. dreggnet-bridge: Watches Dregg journal cells for funded execution-lease requests, dispatches them to dreggnet-exec, and submits signed execution receipts back to the journal.
  4. dreggnet-control: Userspace boringtun WireGuard mesh controller and peer router.
  5. dreggnet-gateway: Fly.io-compatible HTTP/TLS edge gateway routing into mesh workers.
  6. dreggnet-provider & dreggnet-webapp: Provider daemon and operator dashboard.

7.4 /Users/ember/dev/DreggCloud (88 Commits): The House / World / Room Sovereign Firmament

/Users/ember/dev/DreggCloud (88 commits, HEAD e9ff17a, August 29, 2026; README.md, REORIENT.md, HORIZONLOG.md) is the sovereign hosting firmament organized around the House / World / Room ontology in crates/dreggcloud-house:

  • House: The operator's cryptographic custody root, append-only transition journal (Through House Schema v16), and EmitEvent commitment anchor. Every owner lifecycle command is signed with Cipherclerk hybrid Ed25519 + ML-DSA keys (with zero-domain owner signatures strictly refused as of 7a77652).
  • World: A mounted semantic world image (a Dregg store, a Universe dungeon world, or a prepaid service archive) admitted into House custody by exact content digest.
  • Room: An isolated capability workspace or live Sandstorm grain mounted inside a World, rendered to browsers via DEOS Shadow DOM view trees (dc-delivery-*).

Three major custody and serving welds landed at the top of DreggCloud:

  1. House Schema v14 — Live MPC-TLS Evidence (notary-selfhost & notary-live-client, 8ee7127):
  2. Mounts live TLSNotary MPC-TLS attestations (verifying GitHub repository ownership and AWS Bedrock model responses) directly into House custody transitions.
  3. House Schema v15 — Exact Universe World Custody (universe-house-custody/src/lib.rs, 2616663):
  4. Defines AdmitUniverseWorldV1 and ExactUniverseWorldArtifactV1, binding a deterministic dungeon-on-dregg Universe world snapshot (world_id, manifest_sha256, state_root, tick_height) into the House journal.
  5. House Schema v16 — Prepaid Service Archive Custody (service-house-custody/src/lib.rs, e9ff17a & 6aaee3b):
  6. Defines AdmitServiceArchiveV1 and ExactServiceArchiveArtifactV1, mounting prepaid compute/storage service archives into House schema 16 and deterministically reopening remaining prepaid capacity across daemon restarts via offline audit verification.
  7. grain-serve — The sandstorm-serve Capability Weld (grain-serve/src/lib.rs, 31079cb):
  8. Wraps the upstream sandstorm-serve grain launcher behind a House-governed BearerGate and AssuranceRung classifier, ensuring that every HTTP/WebSocket request entering a live Sandstorm grain carries an unforgeable House capability token and records its assurance rung.

8. Closing the Unmerged Lanes: Actionable Engineering Playbook

To bring every open worktree lane across minidregg-lanes, minidregg-wt, nanodregg-wt, delvetalk2-lanes, drorb-swarm-salvage, and DreggCloud to a zero-divergent, fully merged state, execute the following five-stage closure sequence:

8.1 Stage 1: minidregg-lanes (38 Unmerged + 67 Dirty) & minidregg-wt (3 Worktrees)

  1. Land the Five November 2 Consolidation Waves First:
  2. Merge wave-host-light (132b4e178, +14 commits, clean)—which already ports lookupLight, FnEvidence, and FnArchivePublisher into main—and then retire minidregg-wt/fn-evidence (183cd3769) and minidregg-wt/fn-e2-materialized-cache (056846cef).
  3. Merge wave-prose-seed (f81134c8b, +14, clean), wave-apps-workspace (00d88b467, +14, clean), and perf-client (17a7c0811, +12, clean).
  4. Commit or stash the 5 modified files in wave-assurance-detwin (8c42ba8b5, +13, op 151 receipt-opening and certifying-solver), run Assurance/TokenCensus.lean, and merge into main.
  5. Land the Deep Feature Branches (hermes-live, oracle-v7-qualification, app-reds-3, perf-apps):
  6. Rebase hermes-live (048b6b89d, +24, clean) and oracle-v7-qualification (b32f15604, +23, clean) onto the post-wave main tip and run the J14 and Oracle v7 qualification gates.
  7. Merge app-reds-3 (0979c697d, +6, clean: consent worker connection slot queue) and perf-apps (76e22c9d2, +6, clean: doc batch 5).
  8. Harvest High-Value Dirty Worktrees (retained, un-deputy, iter-speed, un-assurance):
  9. In retained (c4ecabe0f, dirty=20 directly on main HEAD), commit the persistent-map served cell cache (store-retained) as a dedicated branch and run read-path benchmarks.
  10. In un-deputy (4f99aa551, +9, dirty=109) and iter-speed (e8a27db8d, +16, dirty=63), separate tracked source diffs from build artifacts before cherry-picking the confused-deputy fixes and Lake build-cache seams.
  11. Prune the 140 clean merged worktrees via git worktree remove and delete the 23 *-target build directories to reclaim disk space.

8.2 Stage 2: nanodregg-wt (9 Worktrees)

  1. Finalize the 5 October 24 Repair Lanes (fixhosting, fixkernel, fixproto, fixrust, fixrust4):
  2. In nanodregg-wt/fixrust4 (234cb4d) and nanodregg-wt/fixrust (b869bb9), squash the WIP card 2+3+4 commits on top of 2d5056e (CLIENT-39) and land via scripts/commit.
  3. In nanodregg-wt/fixhosting (37772b9), finish Hosting/Proofs/FollowAnswer.lean alongside no_refusal_repeats_forever (HOSTING-48), update gates/tcb.tsv, and land via scripts/commit.
  4. In nanodregg-wt/fixproto and nanodregg-wt/fixkernel, commit the Protocol/Proofs/Behind.lean, Protocol/Proofs/Command.lean, and Lang/Journey.lean refinements above 0a299c8.
  5. Archive the 4 Early Layer Worktrees (host, perf, protocol, root):
  6. Extract Host/Proofs/Loop.lean from nanodregg-wt/host and Protocol/{Attempt,Proofs/Attempt}.lean from nanodregg-wt/protocol if still needed for HOST-15, then remove all 4 superseded Oct 16–17 worktrees.

8.3 Stage 3: delvetalk2-lanes (1 Unmerged perf Lane + 44 Clean Merged)

  1. Port 5739d22ec (reachableKnot) from delvetalk2-lanes/perf to delvetalk2 main (fb2179968):
  2. Drop or isolate the scratch commit 996b498a8 (spec/PerfBench.lean) and cherry-pick 5739d22ec6f0bc26e4ac20ad49f0b359e2300032 ("The packet's knot holds only what the entry reaches", touching only spec/Delvetalk/Generics.lean, spec/bend/Compiler/ObjectiveBendElaborate.lean, tests/test_policy.py, and tests/test_turn.py) onto fb2179968.
  3. Run the 576-test suite (python3 -m tests.run) to confirm the 25x packet size reduction (734 KB → 29 KB) and 2x Garden compile speedup at replay edition 20.
  4. Prune the 44 Clean Merged Worktrees:
  5. Run git worktree remove across the 44 fully merged lanes in delvetalk2-lanes while preserving the 5 markdown handoff dossiers (HANDOFF-root-2026-10-14.md, author-build-handoff.md, cx-world-handoff.md, deploy-next-handoff.md, rewrite14-handoff.md).

8.4 Stage 4: drorb-swarm-salvage → drorb

  1. Integrate the 7 Salvaged Lean Modules and Braid-10 Patch:
  2. Copy CacheControl.lean, ContentLocation.lean, RetryAfter.lean, DeployPlus.lean, DeployPlusDrive.lean, Framing.lean, and QuicServer.lean into /Users/ember/dev/drorb/ and apply swarm-tracked-changes.patch to switch drorbServeMeteredBraided and drorbServeBraided in Dataplane.lean from braidedDeployment5 to braidedDeployment10.
  3. Run lake build in /Users/ember/dev/drorb to verify all 1,786 lines of QuicServer.lean (ackRuns_exact, nextGenKeys_preserves_hp, checkRetryToken_sound) and servePipelineOfMetered_braided10_off_eq with zero axioms.

8.5 Stage 5: DreggCloud Upstream Proof-Bridge Closure

  1. Clean Cargo.lock and Complete the Upstream Proof-Bridge Weld (d1a7eef):
  2. Commit or revert the single modified Cargo.lock in /Users/ember/dev/DreggCloud (e9ff17a).
  3. Connect House Schema v16 (service-house-custody) and grain-serve (31079cb) to the newly landed op 151 receipt-opening verifier from minidregg (wave-assurance-detwin), closing the recursive proof-bridge blockers recorded in d1a7eef.