CR-010.R1: freeze the proof spec + ship a conformance suite

[quantamixsol]

CR-010.R1 — Freeze the proof spec + ship a conformance suite (public)

Cherry-pick of the feature commit 1fa41b70 from private PR #318 (merged 2026-07-27, merge commit 9acbf4cf), applied onto public master 356dae49.

What this publishes

GraQle's tamper-evidence proof format is now a frozen, versioned, independently-implementable spec rather than an implicit contract living in code and docstrings.

Added	Purpose
graqle/pct/schema/proof-spec/v1.0/bundle.schema.json	The proof-bundle envelope
graqle/pct/schema/proof-spec/v1.0/keyring.schema.json	The trusted-key manifest
graqle/pct/schema/proof-spec/v1.0/verify-result.schema.json	The --format json result contract
graqle/pct/schema/proof-spec/v1.0/SPEC.md	Prose, verification procedure, threat model
graqle/pct/schema/conformance/	9-case corpus + deterministic golden vectors
tests/test_pct/test_proof_spec_conformance.py	49 tests

Modified — 2 files only: graqle/pct/schema/__init__.py (loader API) and pyproject.toml (wheel artifacts). No signing or verification logic changed.

Why it matters

A third party can now implement a conformant verifier from the published artifacts alone. That claim is tested literally, not asserted: the full corpus runs from an installed wheel at a neutral cwd, over a subprocess/JSON boundary, with no GraQle imports on the consuming side.

• Corpus covers all 7 VerifyFailure members + the exit-2 usage path
• Every expectation is empirically derived from the reference implementation
• Each case asserts the typed failure, the process exit code, and absent-vs-false check semantics
• SPEC_VERSION (1.0) is decoupled from the SDK version — pinning to the spec does not pin you to a release

proof_format_version is published as opaque — deliberately

The field is inside both the ed25519 signed preimage and the leaf hash, so cryptography pins it, not schema validation:

Forgery	Result
Mutate the wrapper value only	UNTRUSTED_KID — signature no longer validates
Mutate wrapper and record	TAMPERED_LEAF — leaf hash no longer matches

A schema enum would therefore add no security while breaking legitimately divergent values already in circulation. Both attacks are pinned as regression tests; the rationale is in SPEC §3.1, and SPEC §8.2 records that any future unification of this field is signature-breaking and needs a migration path rather than an in-place rewrite.

Verification on this public base

• 568 tests green — test_pct + test_verify + test_tamper_evidence, verifier isolation guard included
• Diff re-verified after cherry-pick: exactly the same 20 files, no unrelated lane code, no version files
• TS-1..TS-4 screen: 0 hits

Governance

Private-first per ABSOLUTE RULE #0: private PR #318 opened, reviewed, and merged first; this public PR carries the feature commit only. Sentinel (2-pass adversarial) APPROVE 89%, 0 blockers — pass 1 blocked on a real installed-wheel path bug that is fixed and regression-pinned, and on two further claims that were refuted by measurement. Plan plan_60634c1a · lesson lesson_20260727T190027.

🤖 Generated with Claude Code

[github-actions]

🛡️ GraQle PR Guardian

💥 Blast Radius: 20 modules affected

Module	Files Changed	Risk Level	Impact Radius
graqle	18	🟠 T3	18
pyproject.toml	1	🟡 T2	1
tests	1	🟢 T1	1

Total blast radius: 20

---

🏛️ Governance Verdict

🚫 FAIL

• T3: 5 file(s) require explicit approval.

---

🔍 SHACL Violations

No SHACL violations detected. ✅

---

🔐 Approval Requirements

This PR requires approval from: T3

• A registered Tech Lead or Governance Admin must approve.

❌ Approval requirement NOT yet satisfied.

---

Metric	Value
Blast Radius	20
Files Analyzed	20
Blocked	5
SHACL Violations	0
Verdict	FAIL

---

_{🔬 Powered by GraQle PR Guardian v0.1.0 · Scan completed 2026-07-27T20:35:19.476493+00:00}