# INTEGRITY — E4-H160

## File inventory

| File | Status | Notes |
|---|---|---|
| `README.md` | final | summary and next test |
| `ORIGINAL_CLAIM.md` | preserved | original hypothesis, unedited |
| `PRIOR_ART.md` | final | includes the unretrieved Ghent paper |
| `PREREGISTRATION.md` | preserved | written before the run |
| `METHODS.md` | final | |
| `RESULTS.md` | final | includes the failed T3 |
| `LIMITATIONS.md` | final | |
| `DATA_PROVENANCE.md` | final | states that no data exist |
| `REPRODUCE.md` | final | |
| `INTEGRITY.md` | this file | |
| `code/h160_investigation_v2.py` | final | the only script whose output is reported |
| `results/h160_results.csv` | generated | 17 rows |
| `results/h160_run_log.txt` | generated | console output of the reported run |

## Discarded versions and why

`code/h160_investigation.py` (first version) was **deleted**. It is not preserved because it never
produced a number that reached any document. Its three defects are recorded here so the record shows
what went wrong rather than presenting a clean run that did not happen:

1. **No signal.** The target was placed at 0° while the beam pointed at 20°, so the illumination
   envelope was numerically zero. Every bound came out `nan`.
2. **Zero error artefact.** The delay search grid was symmetric about the true delay, so the true value
   was a grid point and both estimators returned exactly zero error — a fake perfect result.
3. **Aliasing.** A 1 km target at 50 GHz chirp and 4096 samples lies ~81 periods outside the
   unambiguous range `c/(2Δf)`; estimates wrapped to a wrong alias.

All three were caught because the output disagreed with expectation, not because the code crashed.
None of their numbers appear in any reported result.

## Checks performed on the reported run

- Script executes end to end with exit code 0.
- `results/h160_results.csv` contains 17 rows and parses.
- `results/h160_run_log.txt` matches the tables transcribed into `RESULTS.md` (checked by eye).
- Every number in `RESULTS.md` traces to a row in the CSV.
- The claim "24.8×" in `README.md` equals the `ratio_point_ext` value at B = 2000 GHz.

## Checks NOT performed

- No second, independent implementation. The T2 result is a ratio of two bounds computed by the same
  function, so an error in that function would not cancel — but it is also not cross-checked. **This is
  the main integrity gap.**
- No external reproduction by another person.
- No comparison against a published numerical result (none was retrieved).

## Checksums

SHA-256 of the delivered files is recorded in `results/CHECKSUMS.sha256`. Verify with:

```bash
cd Incoprea_E4_H160_2026-11-01
sha256sum -c results/CHECKSUMS.sha256
```

## Claims made and not made

| Statement | Allowed? |
|---|---|
| "The point-target bound saturates and the extended-target bound does not" | **Yes** — computed, in T2 |
| "The floor grows with aperture" | **Yes** — computed, in T1 |
| "The limit is not fundamental" | **Yes**, with the caveat that the extended-target model is idealised |
| "Estimators attain the bound" | **No** — T3 returned 0.75, below 1 |
| "This is new" | **No** — prior art unverified |
| "Experimentally confirmed" | **No** — nothing was measured |