Proof Level Definitions
| Level | Meaning | Buyer-safe? |
|---|---|---|
| Shipped | Released in SB; structural hook/unit tests or catalog invariants prove the contract. | Yes — for the scoped claim. |
| Live E2E (proven) | Canonical host ledger 22/22 with strict-clean methodology; may still carry evidence gaps. | Yes — per host, with gap footnotes. |
| Live E2E (partial) | Some live matrix rows pass; full 22/22 strict-clean not complete on that host. | Qualify by host and row scope. |
| In progress | Design + partial implementation; catalog may ahead of runtime. | No — treat as roadmap. |
| Planned / future | Vision or design doc only; no release-grade proof. | No. |
Claims hygiene: Homepage claims map to
docs/testing/claims-registry.json. This status page is the customer-facing mirror — when they diverge, fix the registry first.Capability Status
| Capability | Proof level | Notes |
|---|---|---|
| APO catalog (27 AF / 22 WF) | Shipped | validate-apo-catalog, generated contract views; 3 process packs (PP-SB-DEFAULT, regulated, startup-fast) |
| Parent-only orchestrator (tier 2+) | Shipped | Directive guard, parent-guard hook tests; marketplace stubs for orchestrator/orient/execute/ship |
| Twelve enforcement layers + V-loops | Shipped | Hook unit suite; completion audit |
| Autonomous session mode | Live E2E (partial) | Proven on fixture rows (OUT-AUTO-01); autonomous_safe/internal_dogfood auto-set mode on tier 2+ at SessionStart |
| Enterprise policy profiles (config) | Shipped | Schema + migration; SessionStart auto-sets mode when tier-safe; supervised/regulated stay interactive |
| 22-row enterprise E2E matrix | Live E2E (partial) | Cursor + Codex canonical 22/22 (live_e2e_proven); Claude 6/22 smoke — tri-host parity incomplete |
| Parallel DAG scheduler | In progress | Catalog + fixture runtime: batch planning, dispatch metadata, join gates, stop-check; flow_queue defers until batch joins; host parallel Task dispatch partial |
| Policy profile runtime enforcement | Live E2E (partial) | clarify_auto, evidence strict, deploy gates partially wired; session mode auto-default shipped |
| Generated certification artifact | Shipped | CERTIFICATION-STATUS.json + site/help/data/certification-status.json; this page hydrates from site copy |
| SOC2-style evidence export | Shipped | scripts/export-enterprise-evidence.sh bundles audit entities + composition log + intent ledger excerpts |
| Durable orchestration event log | Planned | Replay/retry/saga substrate not shipped |
| Cold-start autonomy (vague prompt → shipped app) | Live E2E (partial) | Improving on Cursor/Codex tracks; early rounds needed heavy operator intervention |
| Production deploy without human gate | Planned | Explicitly not offered; policy keeps prod human-gated |
Tri-Host Certification Snapshot
Release-grade enterprise autonomy requires reproducible strict-clean evidence on Claude Code, Codex, and Cursor — not structural tests alone.
| Host | Proof level | Evidence surface |
|---|---|---|
| Cursor | Live E2E (proven) | 22/22 ledger; narrative matrix reconcile; evidence gaps: strict-clean narrative-only, legacy ledger format |
| Codex | Live E2E (proven) | 22/22 ledger reconcile COMPLETE; gates closed_pass; product audit pass |
| Claude Code | Live E2E (partial) | 6/22 registry smoke; gates in progress; strict-clean not eligible |
Canonical: .planning/enterprise-e2e/CERTIFICATION-STATUS.json. Site mirror: site/help/data/certification-status.json. public_autonomous_enterprise_claim_ready: false until Claude reaches 22/22. Operational detail lives in .planning/enterprise-e2e/ (internal).
Known Gaps (Do Not Overclaim)
- Sequential runtime vs DAG catalog — batch planning and join gates exist; full DAG parity and host parallel Task dispatch still in flight.
- Scheduler runtime gaps — batch_dispatch, stop-check, and flow_queue join gate have fixture tests; production skill↔atom matching relies on
migration_map(slug-derived skills may miss joins). - Default interactive onboarding — supervised/regulated profiles still default interactive; autonomous profiles auto-set only on tier 2+ hosts.
- Monitor/ledger drift — enterprise E2E effectiveness plan notes reconciliation must stay non-optional.
- Marketplace command stubs — high-value atomic routes (orchestrator, orient, execute, ship) now have stubs; ~45 skills remain skill-tool-only.
Deeper engineering gap analysis: public gaps index (evidence-backed, not marketing).
Evidence and Export Surfaces
docs/evidence-schema.md— hook evidence record shapesdocs/apo-catalog.json—evidence_recordsand flow-step V-loops.planning/orchestrator-composition-log.jsonl— autonomous composition auditdocs/testing/outcome-criteria-registry.json— blocking autonomy outcomes (OUT-AUTO-01, OUT-ORCH-01, …)scripts/regenerate-site-certification-status.sh— sync certification JSON to GitHub Pages data pathscripts/export-enterprise-evidence.sh— SOC2-style audit zip scaffold (evidence_records, composition log, intent ledger excerpts)
Buyer-facing export UI is not shipped — run the export script locally or in CI for audit bundles.