# Notes

`sh code/run` ran in the image built from `env/Dockerfile` (pinned python:3.12-slim), with no network, in about 1 s; every declared result for C1 matched within its tolerance, and `validation.all_passed` is true.

## Independent recomputation

I wrote `independent_check.py` (in this folder) apart from the bundle's code. It reads the 85 raw API responses under `data/raw/`, keeps the `actual` values for 2023 to 2025, keeps only UTC days with all 48 half-hours, picks each season's three-hour window (start 00:00 to 21:00 UTC, earliest on ties) and one year-round window from 2023 alone, and evaluates them against the 00:00-03:00 UTC baseline:

```
2023 complete days 361 windows {'DJF': '01:30', 'MAM': '11:00', 'JJA': '11:00', 'SON': '00:30'} global 11:00
2024 days 364 midnight 105.117216 seasonal 99.043040 global 115.030220 seasonal vs midnight 5.778479% global vs midnight -9.430428% worse days 137
2025 days 365 midnight 114.099087 seasonal 99.094977 global 111.413242 seasonal vs midnight 13.150070% global vs midnight 2.353958% worse days 111
```

These match the declared values, the selections in `results/selections.json`, and the day counts.

## Source check

I fetched one of the source requests again from the live Carbon Intensity API (`/intensity/2025-06-03T00:00Z/2025-06-16T00:00Z`): all 625 records match the bundle's copy, so the archived response for that period is unchanged at the source.

## Hazard screen

Public, aggregate national electricity carbon-intensity records under CC BY 4.0, and the author's code. No private data, nothing that could help cause harm, and the data are shared under a license that allows it, with credit. Answer: none.