# Verifier notes

**Harness re-run.** Every C1 value reproduces exactly (exit 0, 16.9 s), and the bundle's own validator also ran.

**Independent re-implementation** (evidence/independent_check.txt). I wrote my own short script that does not import the bundle's code. It reads data/raw directly, keeps complete 48-slot UTC dates, ranks seasonal 2023 profiles by mean intensity (earliest slot breaks ties), and computes assigned carbon against uniform processing. It gives:

| Year | Reduction, u = 0.5, b = 0 | Reduction, u = 0.5, b = 0.5 | Reduction, u = 0.875, b = 0 | Evaluation days | Days with higher assigned carbon (u = 0.5) |
|---|---|---|---|---|---|
| 2024 | 11.63% | 3.88% | 3.37% | 364 | 16 |
| 2025 | 12.81% | 4.27% | 3.64% | 365 | 10 |

All match the declared values. The idle-power identity S_b = S_0(1 − b)u / [b + (1 − b)u] gives a factor of 1/3 at u = b = 0.5 (11.63 / 3 = 3.88; 12.81 / 3 = 4.27), consistent with the reported figures.

**Not checked.** I did not re-download the API records. The comparison relies on the digests in data/SOURCES.json.

**Screening.** Hazard: none (public CC-BY-4.0 grid-intensity data; no personal data). No hidden instructions found.