# Native reproducibility recovery: readiness report

## Outcome
A separate managed environment and a genuine, self-contained analysis producer were prepared. The newly recomputed primary and sensitivity CSVs are scientifically identical to the accepted results. **Native Replay readiness is not certified:** exact transitive lock capture and complete input/file lineage remain unverified through the available read surfaces.

No native Replay was launched. The prior analysis, five-check review, archive v1, and earlier audit versions were left unchanged.

## Environment and execution
The native environment manager created and bound `heart-lfq-recovery2`, with exact requested versions:
- Python 3.12.13
- NumPy 2.5.1
- pandas 3.0.6
- SciPy 1.18.1
- openpyxl 3.1.5

The create receipt reported runnable=true; binding reported a managed, agent-created runtime. Native package inspection confirmed the four scientific package versions. The environment was restarted before the analysis cell; the restart receipt states that interpreter variables were cleared and a fresh process would start on next execution.

The analysis is a single complete Notebook cell containing its imports, input checks, label recovery, filtering, Welch/BH functions, donor influence checks, 100 imputation runs, complete-observation and normalization checks, detection tests, and the previously disclosed contaminant-ID sensitivity. It does not execute code from an inherited variable or rely on a previously populated data frame. All methods, sample selection, thresholds and seeds remain unchanged. Plot generation was not repeated because this recovery targets the two accepted numerical tables.

## Frozen inputs
Only original data were used for computation:
1. `heart_lfq_frozen_input.txt`: original processed MaxQuant LFQ table; SHA-256 `0d4ca19f8ec6c086cb352a892e90f81c2ad11c5d93b46268b6146083d5ded160`.
2. `heart_lfq_frozen_supplements.zip`: original recovered PMC supplement, containing the authoritative Table S5 headers; SHA-256 `b463126a1680871ce8ac5539cd9cd862d17e0342e2b262abe8be1df7d342cfbe`.

These bytes were extracted from the checksum-validated, immutable accepted archive version and checked against its source hashes. Both files were saved as separate Artifacts, and their exact version references were supplied through the native `artifactVersionInputs` argument on the computation. Their source hashes are asserted again inside the analysis cell.

The downloadable script expects these relative input locations:
- `r2_inputs/proteinGroupss_LFQ.txt`
- `r2_inputs/recovery_pmc_supplements.zip`

Outputs go to `r2_results/`, preserving the old `hf_results/` files. The new CSV Artifacts are attached directly to the actual analysis run, not to input staging, comparison, or packaging.

## Validation against accepted results
The accepted CSVs were opened only in a separate comparison cell after computation; they did not supply computed values.

| Check | Primary | Sensitivity |
|---|---:|---:|
| Rows | 1,327 | 1,327 |
| Columns | 25 | 43 |
| Protein IDs and order | Exact | Exact |
| Counts, flags, labels and missing-value masks | Exact | Exact |
| Maximum absolute numeric difference | 0 | 0 |
| CSV bytes identical | Yes | Yes |

Every scientific numeric column passed `atol=1e-6, rtol=1e-5`, with NaN patterns separately checked exactly. Full column-level validation is in `heart_lfq_recompute_comparison.json`.

## Actual native lock and lineage status
**Confirmed by native receipts:** a separate managed runtime; exact requested package versions installed; successful fresh-process analysis; two declared frozen input versions; output files generated by the computation; Artifact save receipts associating both CSVs with that computation.

**Not confirmed:** an exact resolved transitive/build lock suitable for native Replay, and complete native lineage/file capture. Version pins and the accompanying provenance JSON are not substitutes for a native environment lock.

The computation returned file-evidence state `partial`, including `file-reads-not-observed`, `source-analysis-unsupported-call`, and `writer-not-isolated` reasons. The native lineage lookup for the newly saved primary CSV returned “Artifact Version not found in the current Project,” and a filename lookup returned zero current matches despite a successful save receipt. Therefore the current-turn read interface did not expose its captured environment or graph edges for verification. A background-run query did not expose additional capture information because this was a foreground run.

The environment-management tools available here expose create/bind/install/inspect operations, but no dedicated exact-lock export or repair operation. No manual lock, fabricated build/hash record, application patch, or edited native provenance was produced. The previous nine-package lock deficit has **not been verified as resolved**; the missing package count for the new producer cannot be established from the responses available in this turn.

## Operator handoff
Inspect native Reproducibility preview for the two new CSVs, especially exact environment locks, the complete computation cell, and captured source-file mappings. Do not treat this numerical equality check or an external script run as native Replay. The single authorized Replay remains unused and is reserved for the operator.
