Performance
Every LeagueMind pick, graded in public — wins, losses, and pushes included. Withdrawn picks stay visible with the reason, never hidden. No pick is ever cherry-picked out.
No picks have a recorded closing line yet. Once graded picks settle against the market's closing price, the average CLV and beat-the-close rate show up here.
Do higher-trust picks actually win more? Each row is a fixed trust band and the record it produced. “Implied” is a rough reference only — the Trust Score is a quality score, not a win probability.
No decided picks yet, so there's nothing to calibrate. As picks settle, each trust band fills in with its real win rate and ROI here.
- Trust Score is a 0-100 confidence/quality composite (edge, market agreement, line stability, data completeness, liquidity) - it is NOT a win probability. 'implied' = avgTrust/100 is only PLAN.md's rough "do 60%-trust picks win ~60%?" reference line; a large actual-vs-implied gap is expected (a real 3% edge on a -110 bet still wins ~53%, not 85%) and is not by itself evidence of miscalibration.
- The signal that IS meaningful is monotonic rank-ordering: higher-trust buckets should win at a higher rate and return a higher ROI than lower-trust buckets. Read the table top-to-bottom, not against the implied column.
- winRate uses wins / (wins + losses + pushes) - pushes stay in the denominator to match compute_record. Markets with frequent pushes read slightly below a true win probability.
- Per-bucket samples are small in early operation; treat any single bucket's rate as noisy until its n is large.
No graded picks yet. Once today's picks settle, they'll show up here — wins, losses, and pushes alike, each with its price, closing line, and CLV.