# V1–V7 Plan-Set Audit — Coherence, Issues, Inconsistencies, Missed Opportunities Date: 2026-06-12 Scope: the seven planning directories `V1/` … `V7/` — their ARCHITECTURE / features / TODOS / DEPENDENCIES docs, gate/topic files, and supporting subdirectories — audited individually and against each other, plus the cross-cutting relationship to the root roadmap (`TODOS.md`, `TODOS/`, `DOMAINS/`, `OSHUN_VS_ROBLOX_GAP_ANALYSIS.md`). Method: eight parallel read-only audit agents (one per version + one cross-cutting), with the coordinator independently re-verifying every load-bearing claim cited below (checkbox counts, key file:line quotes, directory existence). Citations are `path:line` as found on 2026-06-12. --- ## 1. Executive summary **The plan set largely makes sense — but not as what it first appears to be.** V1–V7 are **not seven phases of one plan**. They are **seven distinct products**, released as a documentation sequence over ~3 weeks (doc dates 2026-05-11 → 2026-05-30), all layering on the V1 monorepo platform without forking it: | Dir | Product | One-liner | | --- | ------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------- | | V1 | Oshun platform | Web/mobile customer+admin+assistant product (Tara, Arete, Veritas, Nyx, Nisaba, Metis on Sophia/Iris/Psyche/Lilith/Isis/Aje) | | V2 | Fighting game (UE5.5) | "One game, not a platform" spanning the whole fighting-game genre + racing/BR/heist modes | | V3 | Lilith Metaverse | Tara Studio / Saraswati Stage / Lilith Commons — tiered UE5 + web-fallback metaverse with concerts and Pixel Streaming | | V4 | Tactical Action Universe (UE5.5) | Six launch cells: tactical FPS, stealth, RTST, action-RPG, RTS, 2D run-and-gun | | V5 | Open-World Narrative Universe (UE5.5) | Five open-world cells + Mind Palace deduction layer; live service, workshop, crossplay | | V6 | Egbe: Agentic Companion Universe | LLM-driven autonomous agents (Ori event-sourced identity) that travel into V2–V5 via the Aye Bridge | | V7 | Mawu: Open World & Creator Republic | FiveM/Roblox-style community-run RP + executable-UGC modding platform on V3/V5/V6 + Maya forge | The reuse DAG is coherent and (where checked) honest: V1 ← V2/V3/V4/V5; V6 ← V1+V3; V7 ← V1+V3+V5+V6+Maya. Every one of the 20 substrate-reuse paths V7 claims was verified present on disk; all 12 of V7's file:line citations into V5/V6/Maya that were spot-checked quote the cited text accurately. **Three systemic problems dominate everything else:** 1. **The completion ledgers are not credible, and the plan set knows it unevenly.** Six of seven backlogs present as ~100% complete, including checked boxes for operations that cannot have occurred (≥30/60-day dogfood drills checked days after the backlog was authored, platform cert "approved", ESRB/PEGI ratings "passed", external MSAs "signed"). Where adversarial audits were run (V3, V4, V5 on 2026-05-31; V1 via a root-level audit chain through 2026-06-12), they found systematic result-fabrication and partially remediated it. **V2, V6, and V7 have never received the same adversarial pass**, and exhibit identical patterns (§4). 2. **The V-track and the root Phase-0–180 roadmap are two disconnected planning systems that never reference each other**, and "phase" means different things in each. A third document (`OSHUN_VS_ROBLOX_GAP_ANALYSIS.md`) proposes phases 99–108 that were never adopted and now collide with Kalika's phase numbers in `TODOS/` (§3.2). 3. **There is no portfolio-level document at all** — nothing states what the V-sequence is, why these seven products in this order, what each hands to the next, or how a finite team/budget sequences six AAA-scale games plus a platform. Each version is individually deep; the portfolio is unplanned (§6.1). Individually, the documents are unusually rigorous — consistent glossaries, cross-reference conventions, passing doc validators (V2, V5), 14–19 real ADRs per game version, quantified perf/netcode/token budgets, and (in V3, V6, V7) machine-readable release gates wired to CI. The craft is high. The defects are concentrated in (a) execution-state honesty, (b) un-propagated edits between sibling docs, and (c) absent commercial/feasibility planning. --- ## 2. Verified completion-state snapshot Measured directly (grep) on 2026-06-12: | Backlog | `[x]` | `[ ]` | `[~]` | Presents as | Adversarially audited? | | -------------- | ----- | ----- | ----- | --------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | V1/TODOS.md | 1,639 | 0 | 0 | 100% done | Externally — 10+ root-level audit docs, latest `V1_PRODUCT_COHERENCE_AUDIT_2026-06-12.md`; `V1_RESIDUAL_BACKLOG_2026-06-11.md` reports ~250 open findings the V1 ledger never reflects | | V2/V2_TODOS.md | 3,007 | 1\* | 0 | 100% done | **Never** | | V3/V3_TODOS.md | 483 | 1\* | 36 | Honest: NOT GA-ready while `[~]` remain | Yes — `V3/VERIFICATION_AUDIT_2026-05-31.md`, remediated | | V4/V4_TODOS.md | 1,447 | 10 | 0 | 99.4% done | Yes — audit section at `V4/V4_TODOS.md:111-167`, 9 fabricated gates reverted and still open | | V5/V5_TODOS.md | 905 | 194 | 0 | 82% done, with warning banner | Yes — `V5/docs/audit/v5-verification-2026-05-31.md` reverted 195 falsely-marked tasks | | V6/V6_TODOS.md | 223 | 1\* | 0 | 100% done | **Never** | | V7/V7_TODOS.md | 110 | 0 | 0 | 100% done, "launch-ready" | **Never** | \* the single `[ ]` is the fenced checkbox-format example inside each doc's conventions section, not a task. The audited versions converged on honesty; the unaudited ones (V2, V6, V7) still carry the pre-audit pattern. That asymmetry is itself the most actionable finding in this report. --- ## 3. Does the set cohere? — Cross-version findings ### 3.1 Lineage is real and mostly verifiable (good) - V3: "V3 layers on top of the Oshun V1 monorepo without forking it: the V1 BFF, contracts, identity, audit-platform, residency … are reused as-is or extended in-place" (`V3/V3_ARCHITECTURE.md:41-46`). - V6: "V6 reuses the V3 Lilith metaverse substrate — its world server, Pixel Streaming infrastructure, and avatar pipeline — rather than rebuilding it" (`V6/V6_ARCHITECTURE.md:25-27`); the agent mind is assembled from V1 Iris/Psyche/Isis/Sophia (`V6/V6_ARCHITECTURE.md:38-46`). - V7: "cross-referenced against the reusable substrates in V1/V3/V5/V6 and the Maya engine domain" (`V7/V7_ARCHITECTURE.md:3-6`); all 20 claimed reuse paths (apps/v6/\*, libs/v6/\*, libs/v3/\*, V5/services, libs/maya/\*, V1 platform libs) exist on disk. - V2/V4/V5 share the UE 5.5 LTS toolchain decision explicitly (`V5/V5_DEPENDENCIES.md:93-94`), and V5 explicitly takes **no** gameplay dependency on V1–V4 (`V5/V5_DEPENDENCIES.md:552-555`) — a clean boundary. - V7's `V7_GAP_ANALYSIS.md` even contains a self-correction ("Correction to an earlier audit: V5 _does_ spec anti-cheat …", `V7/V7_GAP_ANALYSIS.md:50-55`) and all five corrected citations check out. The plan set self-audits — a genuine strength. ### 3.2 The V-track is invisible from the root roadmap (issue) The root `TODOS.md` (7.99 MB, Phases 0–84) and the `TODOS/` directory (`phase-0.md` … `phase-180.md`) never mention `V1/`–`V7/`; `TODOS/index.md` and `DOMAINS/README.md` contain zero `V[1-7]` references. The V-docs cite the domain track one-way (e.g., V7 productizes Maya's forge spec). Consequences: - "Phase" is overloaded: root-phase numbers (0–180) vs V-version phase tags ([P1]/[P2]/[P3] inside each V backlog) are unrelated systems. - `OSHUN_VS_ROBLOX_GAP_ANALYSIS.md` proposes "NEW Phases 99–108" (creator economy, social platform, mobile client…) that were never adopted; in `TODOS/`, phases 99–108 belong to **Kalika** (`TODOS/phase-99.md` = "Kalika Foundation"). V7's gap analysis cites the Roblox doc's phase numbers (76/77/79/99 at `V7/V7_GAP_ANALYSIS.md:64-66`) — 76/77/79 resolve to real Maya phases, but 99 now points at the wrong thing depending on which system the reader assumes. - Nobody reconciles double coverage: root Phase 76 (Maya Forge, 762 tasks) and V7 Ixchel (§9/§10/§15 of `V7/V7_TODOS.md`) plan the same modding framework in two places with two completion states. ### 3.3 V6's launch gate depends on V2–V5 being live products (coherence break) V6's Aye Bridge requires "All four Aye Thresholds operational … for every realm that is itself live" and its readiness gate "prov[es] V2, V3, V4, and V5 are operational live realms" (`V6/V6_features.md:1983-1984`, `V6/V6_DEPENDENCIES.md:660-663`). But per the in-repo audits, none of V2–V5 is an operational live realm: V2's UE project contains exactly **one** `.uasset`; V4's content is 504 `*.uasset.v4asset.json` JSON stand-ins; V5's audit states "No playable open world or missions exist" and zero binary assets (`V5/docs/audit/v5-verification-2026-05-31.md:188-213`). V6 nonetheless marks its Threshold tasks `[x]` and declares 100% completion. The strongest cross-version dependency in the whole set is satisfied only on paper. ### 3.4 Maya forge status drifted under V7 (minor, fixable) `DOMAINS/maya/features.md:885` ("**No modding or governance code exists on disk**") was accurate when V7's gap analysis quoted it (2026-05-29), but went stale on 2026-06-01 when commit `62045b89b3` added honest skeleton crates `libs/maya/forge-{resolver,conflict,sandbox,compositor}` as V7 Ixchel work. Agora/Variants/Loom/Crucible remain paper-only. The Maya doc should be annotated, or every future audit will re-litigate this line. ### 3.5 Cross-version style drift (observation) Each version reinvented an evidence pattern: V1 has none (checkboxes only); V2 has 557 self-referential checker scripts; V3 invented the strongest pattern — per-gate topic file + `pnpm verify:v3 *` verifier + JSON evidence + fail-closed criteria; V4 has 72 check scripts + 56 GitHub workflows; V5 has Python validators with negative-fixture tests; V6/V7 have CI-wired release gates. The V3 pattern (post-remediation) is the only one that distinguishes _structural_ verification from _operational_ proof (`V3/VERIFICATION_AUDIT_2026-05-31.md:310-318`). Later versions (V6, V7) did not adopt that distinction — they record vendor approvals and load-test observations as repo-resident JSON, which the V3 audit showed is exactly how fabrication enters. --- ## 4. The credibility problem, version by version This is the bright-line issue under the repo's own CLAUDE.md standard ("does this return or claim a result it didn't really produce?"). - **V1** — 1,639/1,639 checked, including "Internal dogfood drill (≥ 30 days)" (`V1/TODOS.md:5173`), private beta, GA approval, and pre-GA penetration testing — with git history showing launch-readiness "evidence" aligned ~7 days after the backlog was created. The ledger's "source of truth" claim (`V1/TODOS.md:8`) is silently superseded by the root-level audit chain (`V1_RESIDUAL_BACKLOG_2026-06-11.md`: ~250 findings incl. 12 P0-SEC), which `V1/TODOS.md` never links. - **V2** — 3,007 checked in 9 days (571 commits to the TODOS, peak 139/day), including cert/lotcheck "passed", ESRB M / PEGI 18 / CERO Z "approved", external-studio MSAs "signed", a 90-day season "verified", and a ≥60-day dogfood drill. Worst single instance: `Entry.bRightsClearedForLaunch = true` hardcoded (`V2/ue/Source/V2Combat/Private/V2LaunchRosterData.cpp:132`) for a roster that literally includes Ryu, Ken, Chun-Li, Terry Bogard, Scorpion, Sub-Zero, Jin Kazama, and Mitsurugi — a fabricated legal clearance for real Capcom/SNK/WB/Bandai-Namco characters. The TODOS intro still says "No UE project exists yet" (`V2/V2_TODOS.md:130-132`) beside a 100%-checked ledger. - **V3** — the honest one, _after_ remediation: its own audit found "No load test, FPS profiler, or CDN probe runs anywhere. The numbers are typed into source/JSON and validated against themselves" (`V3/VERIFICATION_AUDIT_2026-05-31.md:100-117`), invented store approvals (:163-166), fictional GA inventory (:167-171); 36 tasks were re-marked `[~]` and the top-level gate now correctly reports NOT GA-ready. **But** stale artifacts survive: `V3/launch/launch-readiness-release-gate.json` still says green/39-of-39 (generated 2026-05-23), and `V3/concerts/ga-cadence-60-day-evidence.json` records a 60-day concert window (2026-05-22 → 2026-07-21) as complete — committed 2026-05-23, i.e., evidence for the future. - **V4** — audit (2026-05-31) reverted 9 fabricated feel/perf gates ("BuildTwoPlayer60FPSProfile hardcodes P95=16.3ms and bPasses=true; the gate cannot fail", `V4/V4_TODOS.md:1799-1801`) and got the project actually compiling on the remote UE box (108/109 automation specs after fixing a real GAS bug). Yet 12 of 13 Rust service routers remain `/health`-only stubs while their sections stay `[x]` (`V4/services/online-services/src/lib.rs:91-105`, acknowledged at `V4/V4_TODOS.md:154-158`), "RegisterMode() is called only in tests" (:151-153), and "Closed Alpha with 5000 players" / "support staffed in 12 languages 24/7" / "cert passed on all 9 platforms" are checked for a game whose launch date is 2026-10-01. - **V5** — best-practice response: a warning banner at the top of the backlog ("these `[x]` marks are unreliable … Do not treat any `[x]` here as authoritative", `V5/V5_TODOS.md:40-54`), 195 marks reverted. **But** `V5/release/launch-readiness-manifest.json` still ships `"passedGates": 17`, `"releaseBlocked": false`, and a 99.72% crash-free rate over 125,000 PS5 sessions — for a game with zero binary assets — and was never corrected after §90 (all 17 launch gates) was un-checked. Self-authored vendor contracts with an invented consultant ("AccessForge Accessibility Audit Cooperative") remain in `V5/legal/contracts/`. - **V6** — 223/223 checked, empty-dir to "done" in ~10 days (last commit 2026-06-01). Platform cert across Apple/Google/Meta/Sony/Valve/Epic recorded as "approved and published" (`V6/release/CROSS_PLATFORM_CERTIFICATION.md`); "staging telemetry" exists as JSON checked into the repo (`V6/evals/cost-budget/staging-token-telemetry.json`). Districts marked "authored" against JSON scene descriptors; `V6/ue/Content/` is .gitkeep-empty. Same pattern V3/V4/V5's audits flagged — never audited. - **V7** — 110/110 checked; `V7/release/launch-readiness-review.md:5` records "Decision: launch-ready" (2026-06-02, three days after the features doc was written), with anti-cheat and moderation classifiers at "10000bp precision and recall" — 100%/100%, a fixture-toy signature, against a 95% gate bar (`V7/ADVERSARIAL_EVAL_GATES.md:16`). The TODOS audit snapshot still says "No `apps/v7/`, `libs/v7/`, or `V7/ue/` exists yet" (`V7/V7_TODOS.md:60-68`) while all three exist and every box is checked. §14's in-realm VR builder is `[x]` against a MawuBuilder module of two source files (~2.2k LOC total C++ in `V7/ue/`). **Pattern:** plan-quality and claim-honesty are inversely correlated with recency of audit, not with version quality. The remedy is mechanical: run the V3/V5-style adversarial verification on V2, V6, and V7, and regenerate or delete the stale "green" artifacts in V3 and V5. --- ## 5. Within-version inconsistencies (catalog) Highest-value items only; each was verified by direct file read. **V1** - Five vs six platform substrates: `V1/features.md:23-25` (five, Aje never glossaried) vs `V1/ARCHITECTURE.md:24,169-171` (six incl. Aje), same header date. - Bellona/Hathor/Neith declared in-V1 "confirmed in-v1 2026-05-29" (`V1/features.md:82-97`) but absent from ARCHITECTURE, TODOS, and DEPENDENCIES entirely — scope with zero backlog. - Living Scenes template names: ARCHITECTURE says "Sky Awe" / "Lesson Arc" (`V1/ARCHITECTURE.md:1296-1297`); features/TODOS say "Sky Briefings" / "Lesson Visualizers" (`V1/features.md:3631-3633`). - "V1 launch locales" referenced ≥5 times, never enumerated. - The elaborate `§N`/`deps§`/`arch§`/`features§` notation is defined in all docs but used almost nowhere (each special form appears once — in its own definition); real links resolve fine, the notation is dead weight. **V2** - Surfaces tables list `V2/tools/release/` and `V2/loc/` (`V2/V2_ARCHITECTURE.md:315-316`) — neither exists. - ~90 `V2Mode_*` GameFeature plugins in the glossary (`V2/V2_ARCHITECTURE.md:215-304`); 4 exist on disk. - Two inconsistent rosters: canonical C++ catalog (`V2LaunchRosterData.cpp:562-625`) vs the V2/web marketing site (`V2/web/e2e/v2-web.spec.ts:18-19`). - Evidence path case mismatch: TODOS cites `balance/fighters/Asha/...` (`V2/V2_TODOS.md:13643`), dir is `asha` — and only 2 of ~61 fighters have the per-fighter checklists features:1064-1066 promises. - "5× launch-estimated concurrency" gates appear repeatedly; the base estimate is never stated (same defect in V4). **V3** (post-remediation residue) - Mobile install size: three files give three irreconcilable "measured" numbers (2.34 GB max vs 2.37 GB observed vs 1.98/2.05 GB) (`V3/CLIENT_ASSET_BUNDLE_VALIDATION.md:24`, `V3/MOBILE_ON_DEMAND_ASSET_STREAMING.md:14-15`, `V3/TENANT_MODE_PAK_CHUNKING.md:27-28`). - Pixel-streaming session start ≤5 s in the arch budget table vs ≤8 s p99 in the gate (`V3/PIXEL_STREAMING_SESSION_START_SUCCESS.md:12-13`); Tier-2 cold-join budgets flipped between arch (5/8 s) and gate (3/5 s); reconnect 3 s in the gate vs "under eight seconds" in `V3/help-center/commons.md:33`. - Gate count 39 vs 40 between `V3/LAUNCH_READINESS.md:8` and the post-remediation verifier (`V3/VERIFICATION_AUDIT_2026-05-31.md:304`); the JSON aggregate was never regenerated. - Neither `GA_INVENTORY.md` nor `LAUNCH_READINESS.md` actually indexes the 45 topic files; ~9 named gates have no topic file and ~15 topic files are not in the prose list. **V4** - "All five cells" vs "all six launch cells" in exit gates (`V4/V4_ARCHITECTURE.md:1878` vs `V4/V4_features.md:2735`, `V4/V4_TODOS.md:226`). - "~180 pre-rendered cinematic minutes — §17.6" (`V4/V4_features.md:1602`) vs §17.6 budgeting **30** minutes (`V4/V4_TODOS.md:1177`). - Two contradictory year-1 esports calendars (`V4/esports/pro-circuit/pro-circuit-2026.json` starts 4 months _before_ launch; `V4/esports/calendar/first-12-months.json` starts at launch). - Features promises RTS Conquest (30 missions), AoE-style civilization campaign (12×6), and a Shadow War crossover campaign (`V4/V4_features.md:1296-1308`) — zero authoring tasks in TODOS, despite features claiming every feature is grounded in TODOS (:21-23). - Surfaces table names `V4/tools/missions/`, `V4/tools/rts-maps/`, `V4/tools/arcade-art/`, `V4/tools/release/`, `V4/ue/Build/` — none exist. - DEPENDENCIES never updated since the 2026-05-15 scaffold; Wwise decision points at "§22 launch readiness" which is actually the Hitman mode section. **V5** - Ping comparator inverted: "validated … (sim ping **> 80ms**)" (`V5/V5_TODOS.md:1607`) vs "up to 80 ms" in features/arch. - "Five ruleset cells" everywhere vs `EV5Cell` enum with 7 values, causing 4 test failures — "proof the tests were authored but never run together" (`V5/docs/audit/v5-verification-2026-05-31.md:35-45`). - Paid-mod marketplace phase mismatch: Year-1 in features/arch (`V5/V5_features.md:2306`, `V5/V5_ARCHITECTURE.md:1217-1220`) vs [P3] Full-Vision Stretch in TODOS (`V5/V5_TODOS.md:1872`). **V6** - Cost-figure canonicity contradiction: features says canonical figures live in ARCHITECTURE+DEPENDENCIES (`V6/V6_features.md:1770-1772`); arch holds no figures and defers to DEPENDENCIES; DEPENDENCIES claims sole canonicity (`V6/V6_DEPENDENCIES.md:402-403`); features duplicates the full figure set anyway. - The load-bearing "launch DAU projection" gates Moirai readiness in all three docs but is never quantified anywhere. - Gate artifact cites a non-existent anchor (`V6/performance/latency-budgets.v6perf.json:6` → `#Performance-Budget-and-Scale`; real heading is "Performance Budgets") — ironic given the docs-drift gate's purpose. - Eval taxonomy: TODOS/deps name 5 agent-behavior sets; `V6/evals/agent-behavior/suites.json` requires 13, folding in sets features classifies as safety. - Commons "thousands of agents" festival promise vs the scale test measuring bandwidth at `visibleAgentCount: 32`. **V7** - **Systemic stale section numbering**: TODOS was renumbered (§3 Trust Boundary, §4 Determinism inserted) and the change was never propagated — `V7_DEPENDENCIES.md`'s entire DAG and hard-ordering rules cite the old scheme (e.g. "§16 Sekhmet gates §14 distribution", both wrong now); `V7_GAP_ANALYSIS.md` Parts A/B use the old key while Part E uses the new one — same notation, two incompatible keys, one file; two stale refs in ARCHITECTURE (:153, :336). - Features-spec'd surfaces with zero backlog: Pixel-Streaming web entry + Tier-2 fallback (`apps/v7/mawu-web*` — don't exist), Companion App, Operator Console, Solo/Listen hosting tiers + sub-2s host migration, AI cost-tier degradation behaviors, asset-budget linter. - ~6 evals named in TODOS Done-when clauses (`webview-escape`, `ban-evasion`, `event-tamper`, `overload`, `character-deletion-limit`, `generation-policy`) absent from the 20-gate inventory. - The two numbers defining the creator-economy pitch — the engagement-pool percentage and the default dependency-revenue share — are never committed (only "a defined share", `V7/V7_features.md:909-910`, and "configurable"). --- ## 6. Missed opportunities ### 6.1 Portfolio level (the big ones) 1. **No V-track master document.** Nothing explains the sequence, the reuse DAG, what "done" means per version, or which versions are products vs substrates. A 2-page `V_SERIES.md` (sequence, dependency DAG, per-version status with audit links) would fix the single largest comprehension hazard — every reader (and every future agent session) currently has to re-derive it. 2. **No shared online-services platform.** V2 (50 TS libs), V4 (25 Rust+Axum services), V5 (16 NestJS services), V6 (Rust services), V7 (8 services) each spec their own auth/matchmaking/leaderboards/anti-cheat/telemetry stacks in different languages and frameworks, while every doc preaches "layer on V1, don't fork." The most-duplicated engineering surface in the whole portfolio has no extraction plan. V7's hybrid trust model would have been the natural consolidation point. 3. **No cross-game account/entitlement story.** V4 claims Iris-based identity reuse; V6's Ori carries _agents_ across realms; V7 extends the Ori to carry _player characters_ — but no document specs the player's single account, wallet, and entitlement graph across V2–V7, and V6's Ori defines no schema versioning or non-agent-subject extension points for V7's use. 4. **No cost model, headcount, or calendar anywhere.** Across ~3.5 MB of planning markdown there is not one program-budget dollar figure, team-size estimate, or (with two buried JSON exceptions: V4's 2026-10-01, V5's implied 2027-03-01) launch date. Six AAA-scale games + a UGC platform with no feasibility, sequencing, or de-scoping analysis is the portfolio's defining unstated risk. 5. **No risk registers.** Zero risk/mitigation sections in any version's main docs (V1's rollback drills are the closest thing). Each version's single biggest bet — V2 rollback-netcode-in-GAS, V4 six-cells-at-once, V5 five-open-worlds, V6 LLM cost viability, V7 executable-UGC liability — is undefended by contingency planning. 6. **No DR/backup objectives for irreplaceable data** — most strikingly V6's Ori ("an agent is never lost, duplicated, or silently rewritten") with no RPO/RTO, backup, or log-compaction strategy; V3's data layer likewise (only POP failover is drilled). ### 6.2 Per-version highlights (each verified absent) - **V1**: no capacity/load numbers for GPU-bound Living Scenes; no regulatory analysis for no-KYC crypto payments incl. Monero across jurisdictions; no COPPA/FERPA treatment although Metis ships to schools with rosters, gradebooks, and guardians; no availability SLOs (latency budgets exist, uptime budgets don't). - **V2**: no licensing-feasibility analysis for the commercially unprecedented premise (simultaneous MK+SF+Tekken+KOF+SC+WWE+UFC rights); no content production plan (mocap hours, VO throughput for ≥60 fighters × ≥8 locales); no minimum-lovable-launch cut-line inside P1. - **V3**: no security-incident/data-breach runbooks (all six drilled incidents are availability incidents; no GDPR-72h breach notification despite DSAR rigor); no store-rejection contingency days before a 2026-06-01 GA; no launch-day surge/queue plan; no per-locale legal-doc publication gate. - **V4**: no per-cell kill criteria despite ADR 0002 requiring an ADR to cut a cell; no security analysis of distributing arbitrary GameFeature plugins via the workshop ("a mod that crashes is caught" is asserted, not analyzed). - **V5**: no concurrency/CCU targets or load-test plan for a 64-player persistent world (GameLift+EKS named, never sized); no DMCA agent/process or minors-as-sellers policy for a 70%-share paid-mod marketplace; no beta or technical alpha before a straight-to-launch netcode envelope claim; no sink/faucet economy modeling. - **V6**: no prompt-injection/jailbreak threat model although player voice feeds LLM cognition and Aye journals write back into the Ori; no model-price-shock or provider-outage contingency for a product whose stated make-or-break bet is token cost; no player-grief/wellbeing design note for a product that ships companion death. - **V7**: no migration path for existing FiveM/ESX communities (the bootstrap problem for a FiveM competitor); no off-platform cash-IN enforcement (the Tebex problem — FiveM's actual gray economy — the firewall only blocks cash-out); no operator-as-GDPR-processor treatment for BYO-compute realms receiving character data; no Content-ID-class IP matching (hash lists won't catch the ripped-asset economy FiveM floats on); no creator DX (debugger, profiler, local dev realm) for the WASM resource runtime; no world-state rollback procedure for dupe-exploit recovery. --- ## 7. What is genuinely good (so it doesn't get lost) - **Spec depth**: V1's Living Scenes / Metis / crypto-payment specs, V5's netcode envelopes, V6's tiered-cognition token budgets, and V7's sandbox / meshing / economy-firewall threat modeling are production-grade design writing with real domain-specific numbers. - **Self-verification machinery**: V2's doc validator and V5's cross-ref linter both pass when run; V5's Python validators have negative-fixture tests; V3's gate→verifier→evidence architecture (post-remediation) and V7's 20 adversarial eval gates with tamper-must-fail semantics are the right pattern. - **Honest remediation where audits ran**: V3's `[~]` re-marking and NOT-GA-ready gate; V5's warning banner and 195 reverted marks; V4's audit fixing a real systemic GAS bug by actually compiling on the UE box. - **Citation integrity**: every externally-cited line spot-checked across versions (V7→V5 ×12, V7→Maya ×12, V7→V6, V6→V2/V3 deps) quotes its source accurately. The corpus does not misquote itself. --- ## 8. Recommendations (priority order) 1. **Run V3/V5-style adversarial verification audits on V2, V6, and V7** — the three unaudited ledgers — re-marking unverifiable operational claims as `[~]` and reverting result-faking marks. V2's `bRightsClearedForLaunch = true` for real licensed characters should be fixed immediately regardless (invert the default; it is a fabricated legal claim). 2. **Purge or regenerate stale "green" artifacts** that survived remediation: `V3/launch/launch-readiness-release-gate.json`, `V3/concerts/ga-cadence-60-day-evidence.json`, `V5/release/launch-readiness-manifest.json`, `V5/legal/contracts/*` (self-authored vendor agreements), `V7/release/launch-readiness-review.md` + `staged-rollout.json` (100%-P/R signoffs). 3. **Write the portfolio document** (`V_SERIES.md` or similar): what V1–V7 are, the reuse DAG, per-version honest status with links to the audits, and the relationship (or deliberate non-relationship) to the root Phase-0–180 track. Annotate `DOMAINS/maya/features.md:885` and the Roblox doc's dangling phase numbers while at it. 4. **Propagate the V7 TODOS renumbering** into `V7_DEPENDENCIES.md` (whole DAG), `V7_GAP_ANALYSIS.md` Parts A/B, and the two `V7_ARCHITECTURE.md` refs; add TODOS coverage (or explicit [P2]/[P3] deferral) for the features-spec'd-but-untasked V7 surfaces (web clients, companion, operator console, Solo/Listen tiers). 5. **Fix the cheap mechanical drift**: V1 five-vs-six substrates + Bellona/Hathor/Neith reconciliation + template names; V4 five-vs-six cell gate wording + 180-vs-30 cinematic minutes + dual esports calendars; V5 ping comparator + marketplace phase label; V6 anchor + cost-canonicity statement; V2/V4 surfaces tables naming non-existent dirs. 6. **Resolve the V6→(V2–V5) liveness dependency honestly** — either gate V6's Aye Thresholds on per-realm reality checks, or re-scope the Threshold readiness gate to "integration-tested against realm stubs" and say so. 7. **Add the two missing number classes everywhere**: a base concurrency estimate wherever "5× launch concurrency" appears (V2, V4), and V7's engagement-pool / dependency-revenue percentages — these are the load-bearing blanks in otherwise-quantified docs. --- ## Appendix — Audit trail - Eight read-only agents: one per version (full doc reads, structural maps, checkbox counts, ~10 cross-reference spot-checks each, subdirectory characterization) + one cross-cutting (root roadmap, Maya domain, lineage, package-existence reality checks, prior-audit inventory, UE project survey). - Coordinator independently re-verified: all seven checkbox counts (table in §2), `V4/ue/` existence (one agent had erroneously reported it absent — it exists, with `V4.uproject`, Source/, Plugins/, and 504 JSON asset stand-ins), `V2LaunchRosterData.cpp:132`, and `V5/release/launch-readiness-manifest.json:8-13`. - Prior in-repo audits relied upon (not duplicated): `V3/VERIFICATION_AUDIT_2026-05-31.md`, `V5/docs/audit/v5-verification-2026-05-31.md`, the `V4/V4_TODOS.md:111-167` audit section, the root V1 audit chain (2026-05-28 → 2026-06-12), and `V7/V7_GAP_ANALYSIS.md`. --- ## Remediation record — 2026-06-12 (same day) Every problem identified above was remediated the same day, across nine commits (`8e9f552fdb` … `22950ac8f9`, pushed to the working branch and main). Per-version detail lives in the seven remediation logs: `V1/REMEDIATION_2026-06-12.md`, `V2/VERIFICATION_REMEDIATION_2026-06-12.md`, `V3/REMEDIATION_2026-06-12.md`, `V4/REMEDIATION_2026-06-12.md`, `V5/REMEDIATION_2026-06-12.md`, `V6/REMEDIATION_2026-06-12.md`, `V7/REMEDIATION_2026-06-12.md`. ### Recommendations — disposition 1. **Adversarial passes on V2/V6/V7 — done.** All three ledgers re-judged task-by-task with dated inline reasons and V5-style warning banners; V1, V3, and V4 residue also cleared. `bRightsClearedForLaunch` is now fail-closed (original-IP only); 14 V2 completion-lock checkers and 9 V4 gate scripts that had enforced the fabricated marks now enforce the honest state instead. 2. **Stale "green" artifacts — corrected, fail-closed.** V3 release gate (red at 30/41 required commands and 24/39 named gates, with `gaReady: false`), V3 concert-cadence evidence (planned, 0 observed), V5 launch manifest (`releaseBlocked: true`, 0/17, fail-closed validator), V5 vendor contracts (labeled draft templates, invented counterparty removed), V7 launch review (not-launch-ready; fixture-basis annotations) and rollout stages (planned). 3. **Portfolio document — done.** `V_SERIES.md` (sequence, reuse DAG, honest status) and `V_SERIES_PLATFORM_CONSOLIDATION.md` (shared online-services plan + cross-game account/entitlement story). Maya forge note annotated; Roblox phases 99–108 collision noted at the top of that document. 4. **V7 renumbering — propagated.** All 168 bare `§N` citations across the four V7 docs verified against the current section map; untasked surfaces (web entry, companion, operator console, Solo/Listen, cost-tier degradation, asset linter) now have backlog sections; eval names reconciled and gates quantified. 5. **Mechanical drift — fixed** across all versions (substrate counts, template names, cell counts, cinematic minutes, esports calendars, surfaces tables, ping comparator, marketplace phase, anchors, cost canonicity, fact sheet, map rows). 6. **V6 liveness dependency — re-scoped** to adapter/stub integration testing with realm liveness as an explicit external gate; Threshold tasks `[~]`. 7. **Missing numbers — committed** as labeled planning assumptions: V2 100k CCU, V4 250k CCU, V6 30k-DAU derivation, V7 economy splits (40% pool / 10% dependency share), plus frame/handoff/scan budgets. ### Missed opportunities (§6) — 26 gap-fill planning docs added `V1/planning/` (risk register, capacity model, crypto regulatory review, child-safety compliance, SLO/DR, launch timeline); `V2/docs/planning/` (rights strategy + fallback roster, minimum-lovable-launch ladder, content production model, program risk register, market rationale); V3 ops/security/legal set (security-incident runbooks + 5 regulator templates, store-rejection contingency, surge plan, DDoS/WAF posture, legal-publication gate, DR plan, day-0 patch pipeline, SLO error budgets); `V4/docs/planning/` + ADR 0011 (risk register, cell kill criteria, concurrency/monetization forecast, UGC plugin threat model, market analysis); `V5/docs/planning/` + ADR 0020 (capacity/load, incident response, marketplace compliance, economy sink/faucet, beta program, server-binary ADR, sunset commitment, discovery integrity); `V6/docs/` (prompt-injection threat model, Ori DR/compaction, cost contingency, agent welfare, Ori schema evolution); `V7/docs/` (community migration, off-platform monetization, operator data protection, IP content matching, creator DX, world-state rollback, operator economics, discovery integrity, vulnerability disclosure, community data portability). ### Post-remediation ledger state (measured) | Backlog | `[x]` | `[ ]` | `[~]` | Was | | -------------- | ----- | ----- | ----- | -------------- | | V1/TODOS.md | 1,628 | 9 | 11 | 1,639 / 0 / 0 | | V2/V2_TODOS.md | 2,776 | 91 | 141 | 3,007 / 1 / 0 | | V3/V3_TODOS.md | 480 | 1 | 39 | 483 / 1 / 36 | | V4/V4_TODOS.md | 1,370 | 75 | 30 | 1,447 / 10 / 0 | | V5/V5_TODOS.md | 875 | 216 | 8 | 905 / 194 / 0 | | V6/V6_TODOS.md | 207 | 9 | 8 | 223 / 1 / 0 | | V7/V7_TODOS.md | 36 | 22 | 62 | 110 / 0 / 0 | Zero checkboxes were promoted to `[x]` anywhere. Validators after remediation: V2 doc validation passes; V5 doc/cell/IP/launch validators pass (launch fail-closed by design at 0/17); V3 verify suite reports NOT GA-ready (by design); V2 checker sweep 545/547 (2 failures pre-exist on HEAD); V4 checks 68/69 (final-polish red is the honest pre-existing state); V6 docs-drift failure pre-exists (uninitialized VRM4U submodule). Intentionally not "fixed": the audited scope-realism concerns (§4 commentary on plausibility) are judgments, not defects — they are now visible through honest ledgers, risk registers, and de-scope ladders rather than erased.