Generated capability truth — do not hand-edit status claims. Source:
scripts/audit/capability-truth-registry.json(schemaoshun.capability-truth.v1, as of 2026-08-02). Manual analysis appears only in explicitly labelled Manual commentary fields.
Product status#
| Dir | Product | Scope | Registry status | Evidence | Manual commentary |
|---|---|---|---|---|---|
| V1 | Oshun platform | Customer, admin, assistant, and autonomous content platform | seam-only | Canonical V1 Agentic Studio test count (stale); V1 executed-route enforcement comment (seam-only); V1 autonomy operator browser console (pass); V1 autonomy console live-BFF browser test (pass) | The authenticated autonomy operator route and browser-to-BFF standing-direction path are implemented and tested; the invoked run route still lacks durable complete cost telemetry and the remaining HITL intervention/authority composition. |
| V2 | Fighting game | UE 5.5 fighting/esports product and creator suite | blocked | Current V2 Unreal host preflight (blocked); V2 runtime placement receipt (missing) | The current UE 5.5.4 host is capable but V2's editor build is stale; no V2 execution, cook, playback, or runtime placement receipt is retained. |
| V3 | Lilith Metaverse | Metaverse, concert, stage, and web fallback product | red | V3 external voice-similarity launch task (blocked); V3 launch-readiness release record (red); Current V3 Unreal host preflight (blocked) | The current machine release record is red at 30/41 required commands and 24/39 named gates with GA-ready false. The external voice-similarity gate remains blocked; the current Unreal host is capable but V3 still needs an editor build before qualified execution. |
| V4 | Tactical Action Universe | Six-cell UE 5.5 tactical/action product | blocked | V4 integrated Unreal runtime execution (pass); Current RunPod provider probe (missing) | The integrated V4 Unreal runtime is executed and current; live Calliope/Psyche output, human review, match media, and the composed final-polish release remain blocked. |
| V5 | Open-World Narrative Universe | Five open-world ruleset cells and Mind Palace layer | blocked | V5 Unreal runtime and Linux cook execution (pass); Current RunPod provider probe (missing) | The named V5 systems ran in UE 5.5.4 and real binary assets cooked for Linux; live media providers and signed human release authority remain absent. |
| V6 | Egbe companions | Agentic companions, Ori identity, Orun world, and Aye bridge | seam-only | Current RunPod provider probe (missing); V6 Orun runtime placement receipt (missing) | Provider and runtime-placement evidence is absent even though substantial local cognition code exists. |
| V7 | Mawu creator republic | Server-authoritative RP and executable UGC platform | seam-only | V7 cooked-region runtime placement (pass); Current RunPod provider probe (missing) | The bounded procedural region is cooked and runtime-placed, while headline provider/model generation and human play/canary evidence remain seams. |
| V8 | Ariadne detective universe | Solve-first, self-authoring fair-play detective cases | stale | Canonical V8 test count (stale); Canonical V8 documentation/contract drift verdict (stale) | The last bounded 18-configuration test slice and generated-contract drift gate passed but now exceed their 7-day evidence SLO; neither observation is a production release or live-runtime certification. |
| V9 | Metis learning flagship | Grounded interactive lessons and adaptive learning | blocked | Canonical V9 test count (stale); V9 live-learner release record (missing) | Package/contract tests pass; no live learner-efficacy release record exists. |
| V10 | The Rail | Calm ambient companion shell and cross-portfolio channel programming | blocked | Canonical V10 Rail product test boundary (stale); V10 independent product review (pass); V10 Phase A week-long human exit gate (blocked) | Phase A code and bounded local-runtime paths are implemented and tested; the week-long human exit gate and production content/deployment decisions remain open. |
Status semantics#
implementedrequires clean passing repository evidence but does not mean live or released.liveadditionally requires a passing provider probe and runtime placement receipt.greenadditionally requires a passing release record and no missing, stale, blocked, seam-only, or red evidence.blocked,red,stale,seam-only, andmissingremain non-green states.
Manual commentary#
- V1 through V10 are distinct products layered on the shared platform; their statuses are not phases of one delivery plan.
- A checked source task, test file, or deterministic fixture can support implemented status but cannot establish live or green status without the required probe, placement, and release evidence.
- COV-001 provides the strict canonical content-slot schema. COV-002 provides a deterministic source seed with 28 V1 surfaces, four explicitly incomplete scheduled candidates, V1–V10 launch/open/blocked requirements, and V8/V9 gate/artifact/placement constraints. COV-003 provides the strict portfolio dependency DAG, bounded product views, integrity diagnostics, fail-closed invalidation, and required-frontier scheduler. COV-004 provides a deterministic locale/modality/accessibility/manual-content inventory derived from those targets plus direct source observations, with unknown-preserving quantity aggregates. COV-005 provides strict candidate import/upcast, complete shipped-observation reconciliation, 100% slot/launch/evidence targets, a target-zero unknown-shipped metric, dependency/invalidation blockers, and a typed fail-closed release assertion. The seed still reports zero complete canonical slots, no populated product graph, and no complete product release candidate; product slices still own canonicalization, graph population, current release evidence, artifact accessibility conformance, authoritative shipped observations, and release-path integration.
- RUN-001 provides the strict common Run/Step/Attempt/Artifact/Intervention aggregate. RUN-002 provides hash-chained append/replay, atomic allocation, checkpoint migration, stable-thread cross-process resume, intent-before-effect ordering, archive retention, and a six-point real-process SIGKILL matrix. RUN-003 provides tuple-bound effect keys, a hash-chained intent/claim/result/reconciliation journal, provider-truth-before-retry coordination, conservative keyed/idempotent/reconcile/at-most-once policies, owner-relabel protection, partial-resume state, and worker-lease failover. RUN-004 provides digest-bound, field-accounted one-way adapters for V1 AgentRun plus independent creative cycle/review records, V8 StageJournal plus typed case-gate evidence, V9 Metis orchestration dependencies/governance/evidence, and V6 cognition events plus child authority. RUN-005 now provides one hash-chained submitted-through-withdrawn approval/execution lifecycle within each durable common run record, approval identities bound to plan/source/artifact/policy/gate hashes, automatic stale invalidation, locked revision-checked execution claims, and finalization-before-completion with outputs, failures, rationale, tool/evidence ledgers, and actual cost. Tests reject missing, expired, stale, raced, malformed, and tampered authority and recover a committed final ledger after a rename-boundary fault. HITL-001 now provides an authenticated V1 autonomy operator route with queue/run/artifact/provenance/source/budget/gate/intervention views, complete review and reversal controls, stale-authority warnings, accessible confirmations, optimistic-concurrency recovery, deep browser contract coverage, and a retained browser-to-real-BFF steering proof. This is repository and isolated local-runtime evidence, not an externally hosted production deployment or a provider-backed autonomous cycle. Product databases do not yet consume the common lifecycle, no migration/dual-write/replay/rollback certification has passed, and HITL-002 through HITL-005 still own the complete intervention, transactional co-authoring, notification, and shortcut packages.