Skip to content

formal: prove ceremony algebra and protocol safety kernel#232

Draft
hartsock wants to merge 3 commits into
mainfrom
feat/formal-ceremony-kernel
Draft

formal: prove ceremony algebra and protocol safety kernel#232
hartsock wants to merge 3 commits into
mainfrom
feat/formal-ceremony-kernel

Conversation

@hartsock

@hartsock hartsock commented Jul 16, 2026

Copy link
Copy Markdown
Member

What this PR does

Authored by OpenAI GPT-5 in the Codex desktop harness on Windows.

Test plan

  • git diff --check origin/docs/spec-ceremony-contract..HEAD
  • cargo fmt --all -- --check
  • PR docs(spec): the Ceremony Contract v0.1 draft — five laws, wire objects, chained store #229 is green on Linux, macOS, Python, and Windows at 9e7a245.
  • A normal push reproduced the unchanged no-default Clippy baseline failure: unused Tool and ctx in agent-bridle-tool-shell/tests/unbridle.rs; the force update used the author's previously approved --no-verify fallback only after that reproduction.
  • Review the P1/P2/P0 boundaries and the explicit Tier-1/Tier-2/Tier-3 assumptions before accepting the design.

Out of scope

hartsock added a commit that referenced this pull request Jul 16, 2026
…uite

The monolith (~1025 lines) becomes a dependency-ordered suite of loosely-
coupled, functionally-cohesive profiles — the architecture doctrine applied
to the spec itself, and the synthesis of review-5's partition + GPT-5 #232's
formal kernel.

- README.md — the suite index: dependency DAG, the three-tier proof-teeth
  stack (Tier 1 crypto axioms / Tier 2 protocol Tamarin-ProVerif / Tier 3
  kernel Lean-Aeneas + cross-language conformance vectors), build order
  (P1→P2→P0→{P3,P4,P5}), and the relocated PO ledger. Each profile is a
  DECISION: Proposed → Accepted → Proven, and a downstream decision cannot
  be Accepted until its dependencies are Proven.
- P0 ceremony-contract.md — slimmed to THE NARROW WAIST: five laws, the
  authority lattice, the DecisionSurface seam, gate acceptance, the MITM
  ledger (cross-cutting), governance. 242 lines.
- P1 signed-object-profile.md — CID/canon/Sealed/deterministic-sig/
  allowlist/Profile v1 (foundation).
- P2 chain-store-profile.md — causal-transcript DAG, content/line CID,
  external anti-rollback anchor, the trusted-state-machine model.
- P3 enrollment-protocol.md — Introduction, SAS, anchors; teeth are Tier-2
  symbolic protocol analysis, NOT Lean (a key boundary the reviews implied).
- P4 identity-lifecycle.md — roles/delegation, PinRecord/Audit/Attestation/
  Revocation, break-glass + succession (required, PO-R liveness).
- P5 rendering-security-profile.md — effect binding, gate-signed requests,
  and the honest human-factors residual.

Still five laws; the PO ledger just relocated to the profiles that own each
obligation. GPT-5 #232's kernel is adopted as the P0+P1+P2 provable MVP.

Refs #229, #232, #225, agent-mesh#67.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
@hartsock
hartsock force-pushed the feat/formal-ceremony-kernel branch from c4072ee to 5ceb85b Compare July 16, 2026 02:08
hartsock and others added 3 commits July 15, 2026 23:13
Define the product lattice, trusted protocol state, explicit crypto assumptions, and the Rust-to-Lean refinement gate for the stacked formal-verification work.

Co-authored-by: OpenAI GPT-5 <codex@openai.com>
Split the proof work by the P1 to P2 to P0 dependency order, place rollback state in P2's external anchor, and move enrollment freshness to the Tier-2 protocol track.

Co-Authored-By: OpenAI GPT-5 <codex@openai.com>
Reconcile the formal kernel with PR #229 OB-4 and OB-6: exact envelope recomposition, universal signature domains, named crypto soundness, private verified constructors, and trusted-profile dispatch.

Co-Authored-By: OpenAI GPT-5 <codex@openai.com>
@hartsock
hartsock force-pushed the feat/formal-ceremony-kernel branch from 5ceb85b to f9593d0 Compare July 16, 2026 03:42
hartsock added a commit that referenced this pull request Jul 16, 2026
…urface attestation

The next cut Shawn asked for: turn the tracked OB-1..OB-11 into actual spec
repairs across the suite. All four blockers + the OB-9 soundness bug closed
in text (proofs are the held implementation phase).

Blockers:
- OB-1 dependency inversion: P0 defines abstract AttestEvidence +
  ValidAssociationProof; P4 implements. DAG is now truly acyclic
  (P1→P2→P0→P4→P3; P5 on {P0,P1,P4}).
- OB-2 (author's-call, adopted): linear authority spine per causal thread
  (store_id, thread_id, sequence) — conversation branches, authority is a
  railway; equivocation = two records at one (store,thread,sequence).
- OB-3 attestation transaction: one canonical challenge preimage; CAS commit
  (verify-checkpoint→presence→construct→append→advance-anchor→mint); four
  separate roles (the WebAuthn authenticator is NOT the DAG verifier).
- OB-4 byte carriage: signed-bytes-in-envelope {profile,codec,body,cid,by,
  sig}; JSON/TOML are views, never authority-bearing serializations.
- OB-9 fail-open: resolve(∅,q)=ask (piecewise — a seed would downgrade
  legitimate approve); vocabulary glossary (allow=approve; ask is a verdict;
  escalate is neither).

High/medium: OB-5 reserve→validate→consume-last; OB-6 universal
domain-separation + normative store_id; OB-7 epoch-bound revocation policy +
exact-predicate enrollment; OB-8 conditional PO-R liveness + veto-suppression
threat; OB-10 per-class sealed resource identities; OB-11 suite.toml manifest.

Plus, per Shawn: P5 gains SURFACE ATTESTATION — binary attestation (blessed
measurement, what the surface IS) + rendering attestation (witness-token
ceremony, what it DID) to shrink the faithfulness residual; honestly kept
as shrunk-not-eliminated.

Still five laws. Refs #229 review round 6, GPT-5 #232.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
@hartsock
hartsock changed the base branch from docs/spec-ceremony-contract to main July 17, 2026 00:30
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant