docs(evidence): v0.44 hostile admission audit (#1222, B1.1)

[SisyphusZheng]

Problem

Issue #1222 (Beta.1 hostile admission audit) requires a fresh, independent audit of the v0.44 core before it can freeze as the product baseline: rebuild the semantic-owner matrix from source at the exact candidate SHA, hunt second copies of every named semantic surface, and adjudicate each duplicate-looking implementation against governance constitution §4.3 (canonical owner / reason / parity proof). This PR lands the audit report only — no product code changes (Beta.1 is THIN: audit and report only).

Part of #1222. Stage #1150, umbrella #1155.

Owner

Independent hostile auditor (fresh session, no implementation context), dispatch packet B1.1 under #1222; audit posture per the configured release-verifier agent profile and adjudication authority per docs/governance/GOVERNANCE_CONSTITUTION.md §4.3.

Before

No independent hostile admission audit record existed for the v0.44 core at the Beta.1 candidate SHA; the semantic-ownership registry (docs/current/SEMANTIC_OWNERSHIP.md, self-declared Alpha.9 baseline) had never been cross-checked against a from-scratch source-derived owner matrix.

After

docs/evidence/2026-09-03-v044-hostile-admission-audit.md records, at audit SHA c4b484602b793a1ccc22317fa9e6f2603446b8b4:

• A 33-row semantic-owner matrix rebuilt from packages//tools/ source (§1 of the report), derived before consulting the registry.
• The registry diff (§2): all 29 registry rows confirmed against source; seven divergences enumerated (one unregistered route file-convention surface, four further unregistered surfaces, Alpha.10 citation lag consistent with the declared re-baseline plan).
• A 14-entry duplicate register (§3) with §4.3 verdicts: D1 Part Program validator mirror, D2 VOID_TAGS mirror, D3 when-operator triple evaluation, D8 version literals, D11 SSR stub marker, D12 dual serializers — JUSTIFIED with rerunnable parity proof (all rerun green at the audit SHA); D7/D9/D10/D14 — not duplicates (alias/executor/different contract); D13 — observation (dead legacy markers).
• Three findings reported precisely for thinker issue-tracking (the auditor does not file issues): F1 (§3.3: route file-convention grammar + scanner declaration ordering unregistered), F2 (§4.3: renderer scope predicate mirror rendererScopeMatches ↔ generated __matchingRenderers has no parity proof), F3 (§4.3: duplicate text-node escapeText in both serializers has no named owner and no byte-level text corpus).
• Overall verdict: FAIL (formal) — no behavioral divergence observed anywhere; all parity gates green; admission blocked only on §4.3/§3.3-complete records for F1–F3 (constitution §6.2: a phase that closes with an unresolved §4.3 failure has not closed).

Why-not-second-owner

This change introduces no duplicate-looking implementation: it adds exactly one Markdown evidence document and no code. The report itself adjudicates 14 duplicate-looking implementations in the audited tree; each verdict carries its canonical owner / reason / parity proof per §4.3 (report §3).

Evidence

All at audit SHA c4b484602b793a1ccc22317fa9e6f2603446b8b4 (full command table with exit codes in report §5):

• deno test … packages/element/__tests__/ → exit 0 (246 passed); packages/app/__tests__/ → exit 0 (99 passed, incl. URLPattern native/polyfill corpus); packages/create/__tests__/cli.test.ts → exit 0 (52 passed, version parity).
• Parity guards rerun green: part-program-validation-adversarial (bidirectional validator fault corpus over both program.ts copies), when-operator-convergence, void-tags-convergence, compiled-escape-parity, compiled-element-v1 → exit 0.
• packages/adapter-vite/__tests__/ full suite, full permission set → exit 0 (756 passed, 0 failed, 1 ignored), incl. request-time-parity dev-vs-build (23 steps) and ssg-admission-parity.
• Citation integrity: deno run -A tools/check-audit-citations.ts docs/evidence/2026-09-03-v044-hostile-admission-audit.md → 64 citations, 0 drift.
• Gates on the report commit: deno task fmt:check → 0; deno task lint → 0; deno task docs:truth → 0 (incl. docs:check-version-anchors); deno task docs:check-public/docs:check-strategy/docs:check-current/docs:check-claims → 0; deno task text-integrity:check → 0; deno task docs:check-role-neutral → 0; deno task release:evidence:check → 0; deno task interface:snapshot → 0; deno task package-surface:check → 0.

Scope

One file added: docs/evidence/2026-09-03-v044-hostile-admission-audit.md. Nothing outside the packet was edited; no production code, no registry, no governance text. git status --porcelain confirms the commit contains only this file.

Risk

• F1–F3 require thinker-filed tracking issues before Beta.1 can close per constitution §6.2; the report recommends remediation shapes (registration, binding corpus test, or convergence) but the auditor fixed nothing by design.
• Heavy deployment/browser gates (nitro:proof:*, Playwright suites) were not rerun in this audit; their surfaces were covered by deterministic suites, and Alpha.10 closure recorded them green at 4c3fd116. Noted in report §5.
• An initial gate run without --allow-ffi --allow-sys produced rolldown native-binding failures; root-caused to the permission set, all six affected tests pass with the full task permission set. Recorded in report §5 to prevent misreading.

[github-actions]

APIError: Insufficient Balance

opencode session  |  github run