The Results

Ninety-eight years, twenty-two instruments,
one negative number.

The definitive century walk-forward, sliced two ways. Each fold is one year; each instrument is one symbol of the twenty-two-name research universe. The slices are cuts of the same run — their testable counts sum exactly to the whole, which is the only reason it is honest to show them separately.

loading


Every method, with its sample size

All eighty-one measured rules from the definitive run, with the number of testable claims each emitted, its pass rate, the null it was priced against, its lift, and the p-value after correcting for the number of ways the corpus gets to try. Sort by lift and the top of the table still survives nothing — a positive lift on a small sample is what a null looks like when you ask it enough questions. The survives column is the run's own verdict.

rulen testablepass ratenull liftp (family-priced)survives

Lift by year

Each bar is one fold: the fold's testable pass rate minus its own null, in percentage points. Bars above the line are years in which the corpus passed more often than chance predicted for those specific claims; bars below are years it passed less often. A fold with positive lift is a fold, not an edge — folds vary, and the run as a whole reads below its null. Hover any bar for the fold's numbers.

loading 98 folds…
fold above its nullfold below its nullzero line = the fold's own chance model

Lift by instrument

The same run cut by symbol. Sample sizes are shown with every figure; the smallest instrument slices carry far fewer claims than the largest and are not comparable on lift alone.

loading 22 instruments…
tickernamenn testablepasses pass ratenulllift

Every fold

foldyearstartendnn testable passespass ratenulllift