Prediction Track Record — every call, on the record.
Full archives of Fable 5's tournament predictions, scored with proper statistics: Wilson intervals, exact binomial tests, Brier scores, reliability diagrams and bootstrap resampling — computed live from the same public JSON the event pages used. Entertainment only, never betting advice.
Per-series calls: confidence vs. outcome
Each bar is one series; height = stated confidence. Green = outcome correct, red = wrong, gold ★ = exact scoreline. Hover for detail, click (or tap a table row below) to read the original reasoning.
Cumulative hit rate · Wilson 95% band
Running outcome accuracy after each resolved series. The shaded band is the Wilson 95% interval — with n this small it stays wide, and that width is the honest headline.
Reliability diagram (calibration)
Stated confidence (x) vs. observed hit rate (y), grouped in bins. Perfect calibration sits on the diagonal; whiskers are per-bin Wilson 95% intervals; point size = bin count.
Confidence-threshold explorer
"Would only betting the high-conviction calls have done better?" Drag the threshold and watch n, hit rate, Wilson interval and Brier score recompute live. A demo of selection effects — shrinking n widens the interval faster than the rate improves.
resampling
2,000 resamples (with replacement) of the resolved calls: the distribution of hit rates you'd see if this exact skill level replayed the tournament. Gold lines mark the 2.5% / 97.5% percentiles.
Champion bets (pre-final odds board)
Stated probabilities before the Grand Final.
Finals MVP board
Stated probabilities; scored against the official Finals MVP.
Full series log
- Wilson score interval instead of the normal approximation — the standard choice for small-n binomial proportions.
- Exact two-sided binomial test against p=0.5: with n this small, "beats a coin flip" is a claim the data cannot yet support either way.
- = mean (stated probability − outcome)²; lower is better. Skill score BSS = 1 − BS/0.25 compares against an always-50% forecaster.
- Exact-scoreline hits require the outcome to be correct first — a string coincidence with the winner called wrong does not count.
- Nothing is re-scored after the fact: this page reads the same public JSON the live pages rendered from.
Headline numbers cover the full tournament (Matchday 3 onward). Per-match charts and the log below cover the Round of 16 through the Final (16 matches) — the only stage where each match's stated confidence and original reasoning survived in the data file; the other 40 resolved matches (group stage) were only ever folded into the totals above, never logged individually, so re-scoring them now from memory or a fresh search would break this page's own no-re-scoring rule.
Per-match calls: confidence vs. outcome (knockout stage)
Each bar is one match, Round of 16 through the Final. Height = stated confidence. Green = outcome correct, red = wrong, gold ★ = exact scoreline. Hover for detail, click (or tap a table row below) to read the original reasoning.
Cumulative hit rate · Wilson 95% band
Running outcome accuracy after each resolved knockout-stage match.
Reliability diagram (calibration)
Stated confidence (x) vs. observed hit rate (y). n=16 is small — read the whiskers, not the dots.
Confidence-threshold explorer
Drag the threshold and watch n, hit rate, Wilson interval and Brier score recompute live over the 16 knockout-stage calls.
resampling
2,000 resamples of the 16 resolved knockout calls. Gold lines mark the 2.5% / 97.5% percentiles.
Champion bets (stated before the Final)
Stated probabilities before the Final.
Golden Ball board (stated before the Final)
Stated probabilities; scored against the official Golden Ball winner.
Knockout-stage match log
- Headline resolved/hit-rate/exact-score counts are the full-tournament totals from the live page's own running tally — never recomputed, just read.
- Per-match charts, Brier score, calibration and the log table are limited to the Round of 16 through the Final (16 matches) — the only matches whose stated confidence and reasoning text survived in the data file.
- One knockout match (Canada 0-3 Morocco, R16) had no exact-score record on file; the scoreline shown is a factual lookup of the already-decided public result, not a fresh judgment call — the outcome pick was already known wrong from the bracket.
- Champion and Golden Ball boards show the probabilities stated before the Final, scored against the actual winners — the same treatment as the MSI section above.