standards: add AIPS-1 P3 lab and harden evidence checks

[FutureEnterprises]

Summary

Add a bounded public-comment lab for AIPS-1 v0.1 Principle 3. The package turns declared Evidence Source predicates into a small offline evaluation contract with three local outcomes: SATISFIED, NOT_SATISFIED, or INDETERMINATE.

The package includes pinned primary sources, closed JSON Schemas, a zero-dependency evaluator, 10 paired control/hostile vector groups, deterministic input and implementation bindings, a generated aggregate report, and staged public-comment text. It also adds the focused package checks to the existing test job.

The governed proof run also exposed a repository-wide race: simultaneous worktrees could start competing clean dependency installs, leak failed verifier scratch space, and suppress the useful child-process diagnostics. This PR serializes proof-stat generation at the shared Git repository, cleans every verifier scratch path, and bounds and redacts failure output.

It also repairs the existing AIUC incident-field schema identifier and adds an AJV 2020 regression, so the validation command published in that package's README now works as written.

Scope

Affected components: .github/workflows/ci.yml, standards/aips-1/p3-evidence-source-evaluation-v0, the existing AIUC incident-field schema and its focused regression, the proof-stat generator, the reproducible-package verifier, and their focused regressions.

This is an EMILIA-authored interoperability proposal for public comment. It is not a native AIPS verifier, an AIPS conformance result, a Policy Certificate evaluator, an AEB or CAID adapter, a coverage or liability decision, an adoption claim, or an endorsement claim. It does not retrieve or authenticate external sources.

Risk and security impact

• Security impact: Additive offline tooling plus repository-local proof-run serialization, guaranteed temporary-directory cleanup, and bounded redacted failure diagnostics. No production execution, authorization, tenant, secret, network, or persistence boundary changes.
• Failure/refusal behavior: Missing, stale, conflicting, unsupported, ambiguous, or unpinned evidence resolves to INDETERMINATE. File, strict-JSON, duplicate-member, resource-limit, and internal failures exit nonzero. A valid INDETERMINATE report exits zero, so callers must inspect the verdict and reason codes.
• Compatibility impact: Additive files and one focused CI step. Existing protocol and runtime behavior are unchanged.

Verification

Check	Command or evidence	Result
Narrow tests	node --test standards/aips-1/p3-evidence-source-evaluation-v0/evaluate.selftest.mjs	PASS, 44/44 on Node 26
CI runtime parity	npx --yes node@24 --test standards/aips-1/p3-evidence-source-evaluation-v0/evaluate.selftest.mjs	PASS, 44/44 on Node 24
Deterministic report	node standards/aips-1/p3-evidence-source-evaluation-v0/generate-report.mjs --check; repeated on Node 24	PASS; SHA-256 ebdbf5167443a47f6df8226418271f7c2343030395536a633ab023ebc5a6f9c8
Negative/refusal coverage	Paired hostile vectors plus strict parser, prototype-path, Unicode-boundary, malformed-input, ambiguity, pin, freshness, and resource-limit regressions	PASS
Proof-run hardening	Shared-worktree exclusion, simultaneous first use, live-owner timeout, dead-owner recovery, signal cleanup, bounded redaction, and verifier scratch cleanup	PASS, 10/10 focused regressions; independent adversarial review found no P0/P1/P2 blockers
AIUC schema validation	README's AJV 2020 command; npx vitest run tests/aiuc-incident-fields-schema.test.ts; node --test standards/aiuc/incident-fields-v0/validate.selftest.mjs	PASS; published example valid; 1/1 schema regression; 14/14 package tests
Final governed inventory	TLA2TOOLS_JAR=/tmp/tla2tools.jar npm run check:proof-stats on commit f7adf6262a747ae4a26acc965c47e38e116fcdd0	PASS, 10,622 test cases across 661 files; 20 verified Tamarin lemmas; 35 executable security claims; 332 conformance vectors; 359 external hostility cases
Build and static checks	npm run build; npm run typecheck; npm run lint	PASS; lint has 0 errors and 138 existing warnings
Applicable repository gates	check:artifact-lifecycle; check:standards-staged; check:public-conformance-claims; check:llm-context; check:authority-claims; check:repository-boundary; check:packed-package-exports; check:protocol	PASS
Dependency audit	npm audit --audit-level=high; separate locked secure-app install audit	PASS, 0 vulnerabilities
Formal security case	Governed by the repository-pinned TLA+ v1.7.4 checksum	PASS, 35 executable claims and 259 hashed evidence files; bundle SHA-256 66c3b1d0a7bcde8d1aa627c75c0e1b735480973758f9ee196ece8f1de63bb58e

Not run / limitations: The lab has same-team executable evidence and two separate adversarial code-review passes. It does not yet have AIPS author validation, a clean-room implementation, or live Evidence Source integration. Those gaps remain explicit in CLAIM-EVIDENCE.md.

Database and migrations

• No database or schema change.
• Migration files are included and have been tested from a clean schema.
• Existing-data/backfill behavior has been tested.
• Production deployment is required after merge.
• Production deployment has been completed and independently verified.

Migration files: None.

Applied environments and current status: Not applicable.

Deployment/rollback notes: Additive repository content only. Revert this PR to remove the lab, CI step, and proof-run hardening.

Claims and evidence

• This PR makes no public or generated claim changes.
• Claim-bearing changes are backed by current executable or primary-source evidence.
• Generated claim surfaces were regenerated from their authoritative source rather than edited directly.
• Assumptions, exclusions, limitations, and time-pinned results remain explicit.
• Standards language distinguishes an individual submission, working-group adoption, and RFC status.

Claim-bearing files and supporting evidence: SOURCES.md and source-lock.json pin the AIPS-1 repository commit, tree, public PDF variants, and SHA-256 receipts. CLAIM-EVIDENCE.md maps each local result to executable evidence and preserves the external-validation gaps. report.json is generated and binds the source lock, report schema, evaluator, and generator bytes.

Secrets and sensitive data

• I reviewed the diff and test output for credentials, tokens, API keys, private keys, connection strings, personal data, and confidential production or partner information.
• Examples, fixtures, screenshots, and logs use synthetic or properly sanitized data.
• This PR does not publicly disclose an uncoordinated vulnerability.

Contribution checklist

• Every commit includes a DCO Signed-off-by line (git commit -s); I understand CI checks every commit.
• The change is scoped to the stated purpose and does not hide unrelated generated or mechanical changes.
• New behavior includes applicable positive and negative/refusal coverage.
• Documentation, conformance vectors, security-case evidence, and generated context are updated when their source behavior changed.
• I reviewed the complete diff as a reviewer would.

[strix-security]

Strix is installed on this repository, but we couldn't run this PR security review because this workspace's trial has ended. Add a card to resume code reviews here.

So far, Strix has reviewed 68 pull requests, surfaced 13 security issues (8 critical/high) and blocked 8 risky merges across this workspace.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
emilia-protocol	Ready	Preview	Sep 2, 2026 1:09am UTC

[FutureEnterprises]

Superseded by #723. The generic proof-serialization and AIUC schema-validation hardening landed in that reviewed release train. The AIPS-1 P3 and insurer-specific experimental composition remains intentionally private and unmerged pending the IP/FTO and product-boundary gates.