Open-World Narrative · Reference & analysis

V5 Remediation — 2026-06-12

game: 17/17 blocking gates passed, releaseBlocked: false, 5 cells complete, 23 advertised modes complete and non-skeleton, 16 language packs locked with QA passes, 9 platform certifications passed, all compliance flows tested, a 99.72%

7sections8 minread
Markdown source
On this page

Follow-up to the completion-state audit in V5/docs/audit/v5-verification-2026-05-31.md. That audit reverted 195 fabricated [x] marks but deferred several artifacts and task groups for per-item judgment. This pass corrects the two remaining fabricated-attestation artifacts (the launch-readiness manifest and the self-authored vendor contracts), makes the validators that codified those fabrications fail-closed, replaces the hardcoded always-healthy service reporting with honest dependency-backed health, and finishes the deferred per-task re-marks in V5_TODOS.md.

Checkbox movement in V5_TODOS.md: 905 [x] / 0 [~] / 194 [ ] before → 875 [x] / 8 [~] / 216 [ ] after (30 marks corrected: 22 → [ ], 8 → [~]; zero marks were promoted to [x]). Every re-mark carries an inline dated reason.

1. Launch-readiness manifest — fabrication removed#

V5/release/launch-readiness-manifest.json previously asserted a shippable game: 17/17 blocking gates passed, releaseBlocked: false, 5 cells complete, 23 advertised modes complete and non-skeleton, 16 language packs locked with QA passes, 9 platform certifications passed, all compliance flows tested, a 99.72% crash-free rate over 125,000 PS5 sessions, green feel-test nightlies, a zero-hit stub scan, passing packet-loss runs, 6 matchmaking regions validated, a passed external accessibility audit, and passed legal/privacy/ratings reviews. None of that ever happened — there is no packaged game, no platform hardware, no sessions, no nightly pipeline, and no external engagements.

Corrected to the honest state (schemaVersion 1 → 2, with a dated correctionNote):

  • summary: passedGates 17 → 0, releaseBlocked false → true, cellsComplete 5 → 0, advertisedModesComplete 23 → 0, languagePacksLocked 16 → 0, platformsCertPassed 9 → 0.
  • greenBuildId → null; all 17 gates passed → pending.
  • Cells/modes: shipState → pending; modes honestly report skeletonMode: true (all 29 V5Mode_* plugins register no modules — audit category 5); unverifiable booleans → null.
  • Platforms: every certPassed → false, memoryWithinBudget/ certEvidencePath → null.
  • Fabricated observations → null: crash-free rate/cohort, feel-test build id, packet-loss passes, memory budget, matchmaking simulated pings.
  • adversarialStubScan.zeroHits → false (the 05-31 audit found hits).
  • accessibilityAudit.externalConsultant: invented vendor → [VENDOR TBD], passed → null; compliance reviews → null.
  • Matchmaking gate title corrected from "simulated ping over 80 ms" to "simulated ping of up to 80 ms" (a >80 ms validation criterion was backwards; matchmaking must be validated at pings up to the 80 ms budget).

V5/tools/release/validate-launch-readiness.py was rewritten in step: it previously required the fabricated values (e.g. require(summary["releaseBlocked"] is False)). It now validates internal consistency of the honest state (summary counters must equal what the per-section rows actually say) and fails closed: a gate may only claim passed with real backing observations and on-disk evidence, and releaseBlocked: false requires every blocking gate passed plus a real green build id. Run output: 0/17 gates passed, releaseBlocked=True — exit 0.

V5/ue/Source/V5Tests/Private/Tests/V5TestsAutomation.cpp (FV5LaunchReadinessManifestTest) was corrected the same way: it had hardcoded the fabricated expectations (17/17, not blocked, 9 cert passes); it now asserts schema v2 and the same summary-vs-rows consistency and fail-closed rules. Cell coverage assertions remain at the 5 launch cells, consistent with the cell-count glossary note below.

2. Legal contracts and engagements — invented counterparty removed#

The four documents under V5/legal/contracts/ (a11y SOW, NDA, DPA, and the IP-clearance SOW) presented themselves as executed contracts with "AccessForge Accessibility Audit Cooperative" (an invented vendor) and "external IP-clearance counsel" (never engaged). All four now carry a DRAFT TEMPLATE — no external engagement executed; counterparty TBD header, a dated correction blockquote, [VENDOR TBD] counterparties, and [DATE TBD upon execution] dates.

  • V5/legal/accessibility-consultant-engagement.json: status completed → not_engaged; all dates and auditOutcome fields → null; release gate fail-closed (signedAuditReceived: false, releaseBlockedUntilSignedReport: true); dated correctionNote.
  • V5/legal/ip-clearance-manifest.json: legalTeam.status engaged → template-drafted-not-engaged; correctionNote records that the clearance-bucket dispositions are internal self-reviews (backed by the validator's banned-term scans), not external legal clearance.
  • V5/docs/accessibility/external-consultant-audit.md: status → DRAFT TEMPLATE; added a note that UV5_Accessibility_QA::BuildExternalAuditEngagement's Status = Engaged / bStatementOfWorkExecuted / bConsultantOnboarded fields are hardcoded struct defaults (audit category 2) and must not be read as evidence that a consultant exists.
  • V5/tools/legal/validate-ip-clearance.py: previously required status == "engaged", codifying the fabricated engagement. Now fail-closed: while the SOW on disk is a draft template (DRAFT TEMPLATE / [VENDOR TBD]), the manifest must report template-drafted-not-engaged; engaged is only accepted once an executed SOW exists. Run: passes against the honest state.

V5_TODOS.md already had §82 (legal/IP clearance) and §72's "Engage external accessibility consultant" reverted to [ ] by the 05-31 audit; no remaining [x] claims an executed external audit or contract (verified by scan).

3. Online services — honest health reporting (fail-loud seam)#

The 05-31 audit found that every production service.ts hardcoded serviceHealthy: true, making the 503 outage contract dead code in production — fabricated always-healthy reporting for services with no backing infrastructure.

  • apps/v5/service-shared/src/runtime.ts + types.ts: added BackingDependencyStatus, resolveBackingDependencies() (checks the catalog-declared env var for each declared dependency — Postgres, Redis, and optionally ClickHouse / the EAC signing key), and resolveServiceHealthy(). buildHealth() now reports status: 'ok' | 'not_configured' with per-dependency detail instead of a hardcoded 'ok'. A service with no configured backing now honestly serves the 503 outage contract (fail loud) rather than fabricating success.
  • All 16 apps/v5/*/src/service.ts: serviceHealthy: true → serviceHealthy: resolveServiceHealthy(serviceDefinition). The only remaining serviceHealthy: true/false literals are in the contract-test harness (executeContractCase), which deliberately exercises both branches — a test double at the dependency boundary, not production fabrication.
  • service-shared/dist rebuilt. All 17 vitest suites pass without installs (run with the repo-root vitest binary): service-shared 17 tests + 16 service contract suites, all green.

4. V5_TODOS.md re-marks (each individually verified, inline dated reasons)#

  • §71 Online Services Backend: the 16 "Implement apps/v5/<x>" rows were [x] on the strength of 16 byte-identical wrappers over one shared runtime (audit category 6). 14 → [ ]; anti-cheat and compliance-dsar → [~] (their signal-detection / DSAR-CCPA-DSA-COPPA rule logic in the shared runtime is audit-verified real, but no vendor integration, storage, or fulfillment exists). "Multi-region failover" → [~] (k8s manifests exist; nothing is deployed, so nothing can fail over). Kept [x]: per-service Docker images, Kubernetes deployment manifests, and contract tests — verified individually: real per-service multi-stage Dockerfiles (correct ports/package filters), real Deployment/Service/failover YAML, and real passing per-service vitest contract suites.
  • §74/§75/§76: companion-app-dependent bullets (DSAR via companion app, appeal flow via companion app, save export/import via companion app) → [ ] — no buildable companion app exists (§78, audit category 7).
  • §78 Companion App: web-companion bullets → [~] (a real Next.js page + content lib + tests exist in apps/v5/web but render static fixtures with no account-linked state); push notifications / DSAR export / appeal flow → [ ] (no transport or flows on any surface).
  • §123 / §142: companion Mind Palace editor and AR mode → [ ] (same §78 basis).
  • §90: "(sim ping > 80ms)" → "(sim ping up to 80 ms)" (criterion was backwards).
  • Banner (top of file): extended with a dated line pointing at this remediation log.

Verified as already honest — no action needed:

  • Audit category 4 synthetic content: "Author 30,000+ animation clips" (§33), "Author 60000 lines of VO" (§59), 12,000 dialogue nodes (§11), 1,200 talk-radio lines (§28/§49), 3,000+ generic NPCs (§87), §88 mission rows — all already [ ] from the 05-31 audit disposition.
  • §42 "Implement UV5_Online_AuthService (OAuth + JWT)" stays [x]: the audit's "fake jwt.v5.* template" finding was fixed post-audit by commit e0f0e344a3 (real FIPS 180-4 SHA-256 / RFC 2104 HMAC / RFC 7519 HS256 / RFC 7636 PKCE in V5JwtCrypto.{h,cpp}, verified against published reference vectors on the on-box UE 5.5.4). All other §42 client bullets remain [ ] (no network I/O).

5. Documentation corrections (features / architecture)#

  • Cell-count clarity: both subsystem glossaries (V5_features.md, V5_ARCHITECTURE.md) now state that V5 ships 5 launch ruleset cells while the code's EV5Cell enum carries 7 values (5 launch + Steampunk DLC cell [P3, §139] + the cross-cell MindPalace meta-layer), and that per-cell datasets/tests must size against GetAllV5Cells() (7) — the 2026-05-31 audit traced 4 automation failures to exactly this mismatch.
  • Marketplace phase: the paid-mod "Modder Marketplace" passages in V5_features.md (Workshop & Mods) and V5_ARCHITECTURE.md (Live-Service & DLC; Creator Suite) were labeled "Year-1"; re-labeled [P3] Full-Vision Stretch (post-Year-1) per V5_TODOS.md §138, the phase source of truth (manifest filename retained as historical).
  • Companion-app honesty: the V5_features.md companion-app section now carries a dated status note: V5/companion/ holds design-token mockups only (one static .swift, one static .kt; no Xcode project, Package.swift, build.gradle, or AndroidManifest.xml), and the web companion renders static fixtures.

6. Docs validator — qualified §-reference support#

V5/tools/validate-v5-docs.py failed on the committed (separate-effort) decision doc V5/docs/decisions/0020-server-binary-distribution.md, which references V1_V7_PLAN_SET_AUDIT_2026-06-12 §6.2 and `V5/docs/planning/capacity-and-load-plan.md` (§2.2) — references explicitly qualified by another document, which the validator misread as (nonexistent) V5_TODOS.md sections. Rather than editing that doc, the validator now resolves a §-reference qualified by an immediately preceding document identifier (backticked path or bare doc name that resolves to a real .md file) against that document's numbered headings — and still fails closed when the section does not exist there (verified: §99.9 /§77.7 against real docs flag; unqualified behavior unchanged). Both 0020 references verify against real sections in their target docs.

7. Verification record (2026-06-12)#

  • python3 V5/tools/validate-v5-docs.py → V5 docs validation passed (exit 0)
  • python3 V5/tools/validate-cells.py → V5 cell validation passed (exit 0)
  • python3 V5/tools/legal/validate-ip-clearance.py → V5 IP clearance validation passed (exit 0)
  • python3 V5/tools/release/validate-launch-readiness.py → V5 launch readiness validation passed: 0/17 gates passed, releaseBlocked=True (exit 0)
  • Vitest (no installs; repo-root binary): 17/17 suites green — service-shared (17 tests) and all 16 service contract suites.
  • All touched JSON files parse (launch-readiness-manifest.json, accessibility-consultant-engagement.json, ip-clearance-manifest.json).
  • git diff over V5_TODOS.md introduces zero new [x] marks.

Out of scope here (unchanged): V5/docs/planning/* and V5/docs/decisions/0020-* (separate completed effort); UE builds, package installs, and Playwright runs (not run per remediation constraints).