Wednesday, November 11, 2026

Lab Findings · No. 2

Our headline vs the favorite

Every selection sheet faces one plain question: does its official pick do better than the simplest alternative — taking the morning-line favorite in every race? Our lab ran that comparison for years, and the old record gave a comfortable answer. On the corrected record, the comfortable answer is gone. This post prints both versions, because the change is the finding.

On the old test record, the official book settled 25,654 legs, won 28.1% of them, and returned −16.3% on flat $2 win bets. The morning-line favorite, run over the same record, returned −18.0% on 23,347 legs. The gap between the two books was +1.65 points in our favor. That gap was the headline: a small, steady superiority over the favorite, quoted whenever the sheet’s value was described.

The old record had a defect, covered in the first post in this series: it admitted a race only if the winner already had a prior scored line. That rule deleted 1,528 races before any test ran — every one of them won by a horse with no prior line. A record that removes races based on who won them cannot judge a selection method fairly, because the deleted races are not a random sample. They are races the filed method, which selects horses with lines, was bound to find hard.

The corrected corpus was built from the charts instead: 42,618 chart races reduced to 35,925 thoroughbred races with a unique winner, every starter kept and every exclusion logged with a reason. The headline population used for this comparison carries over the old headline’s own exclusion set, so the old and corrected books face the same kind of races. The only change is that the wrongly deleted races are back in.

With those races restored, the official book settles 26,963 legs, wins 26.87% of them, and returns −20.05%. The favorite returns −19.90% on 24,547 legs. The gap is now −0.15 points — the official book runs a fraction behind the favorite, not ahead of it. The rebuild also ran a machinery check: on the races both records share, the corrected machinery reproduces the old numbers exactly, to the fourth decimal. The move comes from the recovered population alone, not from any change in how the books are computed.

The recovered races explain why. On the 1,309 recovered legs that settled, the official pick won 31 times and the book returned −92.84%. That number looks alarming until its construction is stated: a recovered race is, by definition, a race won by a horse with no prior line, and the filer selects horses that have lines. The favorite did better on the same races and still returned −57.28%, winning 15.25% of them. These races are hard for any method built on past form. The old record did not solve them; it hid them.

The verdict word in the lab record for this family is MOVES, with the note that the superiority is gone. So the sheet makes a narrower promise than it once did. It does not claim an edge over the favorite. It promises a disciplined selection — one pick per race, filed before the races run, with the backup order fixed in advance — and it settles that selection in public, every day, in the scorecard at the top of this post. Yesterday’s line above is what that promise looks like on a losing day: printed in full, not rounded into a better story. A reader who wants the favorite’s record instead can read it here too: on the corrected corpus, the favorite loses at almost exactly the same rate we do, and neither book is a winning one.