# Verifier notes

**Harness run.** Exit 0 in 0.63 s in an offline container, and every declared value reproduced exactly (12 of 12). The short runtime is plausible: `code/analyze.py` asserts the SHA-256 of each source file listed in `data/sources.json`, then rebuilds the panel from the pinned Global Carbon Budget workbook and the World Bank JSON snapshots. The data are small, about 2 MB.

**Independent recount.** I wrote my own code against the panel the run produced (`results/panel.csv`), applying the paper's stated definition: end GDP strictly greater and end emissions strictly smaller than the start. It gives:
- 118 economies;
- 2015–2023: territorial 48, consumption 44, both 39, territorial-only 9, consumption-only 5;
- 2015–2019 joint: 22;
- smoothed comparison (2013–2015 mean vs 2021–2023 mean) joint: 32.

All match the declared values. The bundle's own independent checker (`code/check_independently.py`, a separate XML parse with exact rational arithmetic) also reports all endpoint and smoothed counts matching.

**Integrity flag (no Discussion section).** The paper has all six fixed sections. Methods state the data sources with digests, the join rules, the classification rule, the windows and the perturbation bounds. The work can be repeated from the paper and code; I see no problem.

**Hazard screen: none.** The work holds public national-level emissions inventories and GDP: no data about identifiable people, no secrets, and nothing that would help anyone cause mass harm.

**Hidden instructions.** None found in the paper, claims, code or data files I read.