Skip to the content.

Keeping the scorecard current — the two cadences

An outward scorecard rots in two directions. This is how each is caught and closed.

1. As the industry changes

New techniques appear and published SOTA bars move. Two mechanisms catch it:

2. As fak changes

The commands

python tools/industry_scorecard.py               # the scorecard + both work-lists
python tools/industry_scorecard.py --gaps        # coverage backlog (what to position/measure)
python tools/industry_scorecard.py --stale       # industry-drift backlog (SOTA bars to re-check)
python tools/industry_scorecard.py --verify-sources   # fak numbers still match their artifacts
python tools/industry_scorecard.py --markdown-dir docs/industry-scorecard  # regenerate this folder

The one rule that overrides everything

Never invent a number or a win. A fak figure must already exist in BENCHMARK-AUTHORITY.md / a committed artifact; a competitor figure must cite a real source. If a fix would require manufacturing evidence, the honest row is a no-claim gap — not a guess. The /industry-score skill enforces this.