# Rounds 7 and 8: pre-registered scoring addendum

Fixed on 4 November 2026, after lock (02:10 UTC) and before any question resolved. It changes no locked call.

1. The headline score stays the reconciled two-marker result (markers A and B, reconciliation R): hit rate and Brier score.
2. Added scores, reported for every round from Round 7:
   - Log score and a baseline score on the Metaculus scale (0 for 50% on every binary, 100 for perfect foresight).
   - Expected calibration error over ten equal-width bins.
   - Brier skill against two naive baselines: 50% on every binary, and the base rate recorded in each call.
3. Market comparison: where a Polymarket or Kalshi market matches a question closely enough, its midpoint price as captured on 4 November 2026 is scored alongside our call. A 50/50 blend of ours and the market's is also scored as a shadow. We report the disagreement buckets (gap of 0-10, 10-20 and over 20 points). Shadow tracks never replace the official score.
4. Voids follow the round rules. A void question is removed from every score track alike.
5. Commitments: all 500 salted call hashes are numbered and every one will be revealed, including the bad calls.