# REPRODUCE — E4-H160

## Environment

- Python 3.12
- numpy (only dependency)
- Interpreter used: `/home/toby/Desktop/300 study 1/04_SUPPORTING/.venv/bin/python`
- Standard library: `csv`, `os`

Check:

```bash
python3 -c "import numpy; print(numpy.__version__)"
```

## Run

```bash
cd Incoprea_E4_H160_2026-11-01/code
python3 h160_investigation_v2.py
```

Runtime: ~2–4 minutes (the Monte-Carlo dominates).

## Outputs

| File | Contents |
|---|---|
| `results/h160_results.csv` | 17 rows: T1 (6), T2 (7), T3 (4) |
| `results/h160_run_log.txt` | full console output of the reported run |

## Expected values (for checking a re-run)

Deterministic except for the Monte-Carlo, which is seeded (`np.random.default_rng(20261101)`).
Re-running reproduces T1 and T2 exactly and T3 to within Monte-Carlo scatter (120 trials per point).

Spot checks:

- T1 at N = 2048: `B_eff = 214.0 GHz`, `floor = 700.46 µm`, `floor/D = 0.171`.
- T2 at B = 2000 GHz: `CRB point = 0.0011 ps`, `CRB extended ≈ 0.0000 ps`, ratio 24.78.
- T3 at 100 GHz matched: `RMSE = 0.0069 ps`, `CRB = 0.0092 ps`, ratio 0.75.

If T3 returns a ratio near 4×10⁶ (i.e. an error ≈ 1/Δf), the delay grid has wrapped — check that the
target range is inside `c/(2Δf)`.

## Known non-determinism

Only T3. If exact reproduction is required, the seed makes it deterministic; across platforms the
float summation order in the FFT may shift the last digit.

## What does not reproduce

The `h160_investigation.py` file referenced in earlier working notes was deleted after three inside
the same session it produced invalid output (noise-only signal, grid landing on the true delay, and
delay-axis aliasing). It is not preserved because none of its printed numbers were used. Its three
defects are documented in `INTEGRITY.md` so the failures are not lost.