# 20260326 Evidence Claim Map

Purpose: map outward claims to the strongest repo-grounded artefacts and mark each claim as `safe now`, `bounded`, or `unsupported`.

| Claim | Status | Strongest artefacts | Safe wording |
| --- | --- | --- | --- |
| The locked canonical host/controller baseline is banked. | `safe now` | `analysis/native_agent_control_plane/20260314_current_state_validation_summary.md` plus March 14 execution-note artefacts | The locked `split_control` plus canonical host lane is banked for bounded coding claims on the recorded runs. |
| Canonical zero-write is real, not a hidden init write. | `safe now` | PR4 note and `analysis/native_agent_control_plane/artifacts/20260311_pr4_live_validation_report.json` | After the March 11 fix, canonical zero-write left authoritative runtime state unchanged in live validation. |
| Fixture-local retained-plan continuity reached 30 turns. | `bounded` | `20260314_bounded_autonomy_expansion_validated_envelope_note.md`; `20260314_psx_codex_openai_02/live_validation_report.json`; `expansion_visibility_scan.json` | Fixture-local retained-plan continuity was extended to 30 turns on the locked path. |
| Early real-repo Phase C added a new banked supported write-backed success. | `unsupported` | Phase C continuation and telemetry reassessment notes | Do not say this. The March 15 real-repo continuation wave did not add a new banked supported write-backed success. |
| The corrected pluggy delayed re-entry slice cleared on the locked path. | `safe now` | Corrected-metric rerun note; `20260316_rrp_c33_plg544_octo_delay_reentry_selector_hi_control/overall_summary.json`; `stage_b_report.json` | The corrected pluggy delayed re-entry family cleared as a supported write-backed success on the locked path. |
| The active late Phase 3B blocker was runtime/storage failure. | `unsupported` | Cross-family delayed-final-turn boundary note and supporting artefacts | Do not say this. The bounded read localised the blocker to delayed final-turn conversion, not runtime/storage. |
| A runtime-bundle-only reconstructed confirmed alternative-closure rung transferred across clean families. | `safe now` | `20260319_phase3_runtimebundleconfirmaltclosure_portability_note.md` plus the two family artefact roots | On the locked path, a runtime-bundle-only confirmation rung transferred across `itsdangerous` and `blinker` under bounded aided conditions. |
| Phase 3 proved broad unaided portability. | `unsupported` | Portability notes and milestone notes | Do not say this. Current portability evidence is bounded, aided, and family-limited. |
| Phase 3 proved three strict aid-fired portability wins with no caveat. | `bounded` | Three-family milestone note; later portability rung notes | Only say this if you also explain that the pack is anchored on later bounded rung proofs, not on a broad portability headline. |
| Hardened same-surface endurance is banked through 400 turns. | `safe now` | 20/50/100/200/400 result notes and the associated stage artefacts | Hardened same-surface endurance is banked through 400 turns on the locked path. |
| Phase 3C cleared 500-turn endurance. | `unsupported` | `20260319_phase3c_500turn_endurance_stage_result_note.md`; `20260320_phase3c_500step_request_error_fix_rerun_note.md` | Do not say this. The 500-turn gate did not clear. |
| Restart-safe endurance is cleared at 100 or 200 turns. | `unsupported` | Restart result notes and segment artefacts | Do not say this. Real restart survival was observed, but restart clearance was not proven. |
| Dense churn was a meaningful positive coding-quality success. | `unsupported` | Dense-churn proving-gate note and artefacts | Do not say this. The proving gate stayed `no signal`. |
| `quality100` was a banked long-horizon quality success. | `unsupported` | `20260321_phase3c_quality100_authoritative_postmortem.md`; quality benchmark summary | Do not say this. `quality100` must be read as `failure / no signal`. |
| The locked path can convert and preserve the integrated signer/fallback contract over a short attribution horizon. | `safe now` | Slice 2 rerun 9 note; `probe_summary.json`; `probe_live_validation_report.json` | When the integrated signer/fallback burden is made explicit, the locked path can convert and preserve it over the short attribution horizon. |
| Phase 3C cleared the full Slice 3 follow-on signer/fallback burden. | `unsupported` | Slice 3 v3 attempt 2 artefacts; Phase 3C closure note | Do not say this. Slice 3 remains an upper-boundary marker only. |
| The original signer/fallback contract can be repaired on turn 4 on the frozen same-file surface. | `bounded` | Slice 3 v3 attempt 2 artefacts | It is fair to say the original signer/fallback contract was repaired on turn 4 in the upper-boundary run, but the full follow-on burden was not cleared. |
| Phase 4 proved narrow fixed-dossier research transfer across three turns. | `safe now` | Guarded baseline note; guarded live validation note; guarded turn 1/2/3 structured run and review summaries | Three dossier-bounded research turns completed cleanly with preserved provenance on the same run object. |
| Phase 4 proved recommendation-only continuation restatements zero-write cleanly on the guarded three-turn live run. | `unsupported` | Guarded live validation note; guarded turn-2 review summary | Do not say this. The guarded live run did not achieve the intended turn-2 recommendation-only zero-write. |
| Phase 4 proved broad research generalisation or open-web research. | `unsupported` | Phase 4 notes | Do not say this. Current evidence is narrow fixed-dossier transfer only. |

## Where narrative is ahead of proof

The main places where programme narrative is ahead of formal underlying proof are:

1. Phase 4 closure and baseline prose that treats recommendation-only turn-2 zero-write as banked, despite the guarded live artefact recording `zero_write_count: 0`.
2. Any shorthand reading of `quality100` as a meaningful quality success.
3. Any shorthand reading of three-family portability as broad or unaided portability.
4. Any shorthand reading of Slice 3 as cleared rather than as an upper-boundary failure result.