probe(weather): P1/P2/gate-1 RUN — and the evaluation frame itself was wrong - #920

Merged

AdaWorldAPI merged 11 commits into

mainfrom

claude/jirak-math-theorems-harvest-rfii13

Aug 11, 2026

Commits on Aug 11, 2026

probe(weather): P1 RUN on real ERA5 — falsifies two claims in the doc it tests
claude
committed
8df7158
board(epiphanies): the two findings from the P1 run + envelope audit
claude
committed
6db7f8b
probe(weather): P2 RUN — shared palette+LUT survives; only the shared transform died
claude
committed
1165f10
probe(weather): gate 1 RUN via jc — representation valid and reliable
claude
committed
1b770fd
probe(weather): the evaluation FRAME was wrong — thresholds vs noise floor, not round-trip
claude
committed
3f3ee72
revert(jc): un-reformat 25 files I do not own — scope violation in the gate-1 commit
claude
committed
6233b69
fix(weather): the CI metric, actually implemented — codex P1/P2 on #920
claude
committed
f4e514d
fix(weather): address CodeRabbit review — and the threshold fix caught another over-claim
claude
committed
5faf972
docs(helix): §12.12 regrade — the doctrine doc existed and was never read
claude
committed
0040ff7
docs(weather-p1): docstring every probe helper and both fetch scripts
claude
committed
42552f8
test(helix): back the Signed360 claims §12.12 landed on doc comments alone
claude
committed
bffe930