EVIDENCE_INDEX.md

# Evidence index

Paths beginning with `results/`, `src/`, `benchmarks/`, or `manuscript/sections/` below refer to the extracted **original research archive**. Paths under `evidence/` refer to convenient copies in this Hugging Face kit. Copies are byte-identical to their source records unless explicitly labeled packaging-time output.

| Claim or artifact | Authoritative archive records | Inspection source |
|---|---|---|
| Central orthogonal-task formulas and dual construction | `results/release_summary.json`; `results/scenarios/raw.json` analytic witnesses | `manuscript/sections/morphogenesis_theory.tex`; `benchmarks/run_scenarios.py` |
| 18 scenario solves and closed gaps | `results/scenarios/protocol.json`, `raw.json`, `summary.json` | `src/ailith/scenario.py`; independent checks in `benchmarks/run_scenarios.py` |
| 37 inherited-state schedules, 35 closed gaps | `results/inheritance/protocol.json`, `raw.json`, `summary.json` | `src/ailith/warm_grid.py`; independent checks in `benchmarks/run_inheritance.py` |
| Static reserve design, 13 cases / eight optimizer starts | `results/inheritance/raw.json`, `summary.json` | `src/ailith/morphology.py`; `benchmarks/run_inheritance.py` |
| 71 streaming schedules | `results/streaming/protocol.json`, `raw.json`, `summary.json` | `src/ailith/async_grid.py`; `benchmarks/run_streaming.py` |
| Requested-target and context bindings | `results/streaming/binding_audit.json` | `benchmarks/audit_streaming_bindings.py` |
| 93 scalar warm records and 25 paid pairs | `results/reserve/protocol_default.json`, `raw_default.json`, `summary_default.json` | `src/ailith/reserve.py`; `benchmarks/run_reserve.py` |
| Timed hereditary arena | `results/hereditary_arena.json` | `examples/hereditary_arena.py`; `src/ailith/evidence_stream.py` |
| Frozen 220-test validation | `results/validation.json`, `results/test_log.txt` | `tests/`; `benchmarks/run_validation.py` |
| 79-page PDF and recorded automated page review | `results/review/final_release_qa.json`; `results/release_summary.json` | `manuscript/`; review is AI-assisted layout QA, not human peer review |
| Original source integrity | `MANIFEST.json` | `verify_release.py`: 900 checks of inventory and recorded bindings |

## Convenient summary copies

- `evidence/release_summary.json`
- `evidence/inheritance_summary.json`
- `evidence/scenarios_summary.json`
- `evidence/streaming_summary.json`
- `evidence/reserve_summary.json`
- `evidence/frozen_validation.json`
- `evidence/hereditary_arena.json`

Raw arrays, schedules, dual variables, failed development attempts, and full review records remain in the complete archive. Summary copies do not replace those detailed records.

## Packaging-time evidence

`evidence/packaging_verification.json` records the current extraction, 900 original verifier checks, a fresh 220-test run, byte identity of copied artifacts, environment information, and the source archive hash. `evidence/packaging_regression.txt` is that new test transcript. These outputs are separate from all frozen records. No numerical study was rerun to create this Hugging Face edition.

## Two retained open inherited-state gaps

| Record | Feasible action | Numerical lower bound | Relative grid gap |
|---|---:|---:|---:|
| `matrix_01_n8` | 3.949451727595017 | 3.9126183565936166 | 0.9326198557% |
| `matrix_05_n8` | 22.84641108508443 | 22.841337197898717 | 0.02220868375% |

The release keeps both schedules because their independently checked feasibility remains useful. Their displayed intervals do not establish the exact optimum.

## Evidence levels

The manuscript's analytic statements are supported by their written proofs under explicit hypotheses. Finite-grid numerical bounds are floating-point checks, not proof-assistant or interval-arithmetic certificates. Internal independent oracles are distinct implementations, not external scientific replication. Source hashes identify bytes, not authorship, and passing software tests does not establish learned recursive intelligence growth.