People ask where the method comes from, and the honest answer is a bookshelf. The Daily Pace Card was assembled in a lab — pre-registered tests, full-data backtests, versioned protocol — but the ideas it runs on were not invented here. They were taken, one at a time, from forty years of handicapping literature, and every one of them had to survive a test on our own data before it earned a place in the product. Five sources made it. The rest of the shelf was tested too, and this post names those as well, because a lab that only reports its wins is not a lab.
Charles Carroll’s Handicapping Speed (1991) is the load-bearing book. His track-constant method says a raw time means nothing until it is read against the track that produced it — every circuit, distance and surface carries its own constant — and his projection variant reads a horse’s line against what that same horse has done before. That is, almost exactly, our speed figure. Every line on our sheets is adjusted against the track’s own constants and projected off the horse’s own prior lines. OUR_RATING v4, and the figure printed on every card, descend from Carroll more directly than from anything else we have read.
The Sartin lineage contributed a principle rather than a formula: rate a horse against its own demonstrated lines, not against a class ladder somebody else built. A horse’s best evidence is its own record, projected honestly. That principle sits inside the same figure build as Carroll’s constants — the two ideas are really one idea seen from opposite ends — and it has been live in the product since October 29.
Randy Giles’s Extreme Pace Handicapping (2008) gave us pace pressure — the notion that a race’s shape can be typed by how much early speed it contains and where that speed sits. We tested his race-shape types over 207 races, closed them, and rebuilt the idea as our own six Race Shape types with points-only scoring. The names and the math on the current sheet are ours. The seed was his, and the sheet would not look the way it does without it.
James Quinn’s The Complete Handicapper (2012) asked a prior question: not how to handicap a race, but which races are handicapable. His split measured real on our data — maiden races are the least handicapable population on the card, the fields where form is thinnest and outcomes run widest. That finding entered the selection protocol in v2.6 as the maiden exclusion, and it survives in the cell rules the protocol runs today.
Mark Cramer’s Value Handicapping is here indirectly, and the indirectness is the point. His overlay claim failed its test outright. But building that test forced us to rebuild our own-line figure cleanly, and the rebuild became OUR_LINE v1. Own-line ranking later lost to ML-BASE as the ranking anchor — the morning line still orders our fields — so Cramer’s residue lives in the figure work, not the ranking. A failed test that leaves the machinery better than it found it has still paid rent.
The speed figure and the weight beside each name both descend from Carroll’s track-constant method in Handicapping Speed (1991). Every past line is adjusted against constants estimated for its own track, distance, and surface, then projected off the horse’s own prior lines — the Sartin principle, each horse rated against itself. A horse’s figure is the median of its last three own lines on that scale. The weight is built from the same figure: we solve horses connected by common opponents into a web, and anchor each web so its average horse carries 120 lb. That is why the sheet prints a weight — “125 lb” — and never the raw figure beside it, and why weights from different webs are never compared without naming the web.
No published claim entered the sheet untested, and most did not survive the test. Carroll’s constants and projection passed — they are the figure. Quinn’s handicapability split measured real and shaped the maiden cell rules. Giles’s pace-pressure types were tested over 207 races, then rebuilt as our own six race-shape types; they annotate the card and never rank it. Cramer’s overlay claim failed, and its test produced the clean own-line rebuild. Scott’s APV and class rating lost as rankers; his form screen measured real. Taulbot’s angles, Brohamer’s track model, and Hall’s pace claims were not supported. Beyer’s figures were a cross-check only.
The ranking took the longest to settle, because everything we tried lost to the morning line. Own-line ranking lost to plain morning-line order. A ranking search over the corrected corpus of 35,925 races promoted no candidate. The strongest challenger, a logistic model on market probability, figure, weight, pace, draw, recency, and starts, missed its frozen gates — a holdout paired margin of +2.76 against a +3.00 gate — and stands locked as a paper record. So the ranking is the morning line itself. The top four are scored by consensus on five factors: morning line, our weight, our speed figure, early pace, and draw, taken from a frozen draw table built from the corpus. Our own pick carries five points as a separate column; the highest total is the selection. One limit, stated plainly: a decomposition of the consensus’s economics puts its edge inside its uncertainty interval, so we run it as a selection aid, not a wagering claim.
The rest of the shelf went through the same gates. William Scott’s APV and PCR were rejected as rankers; his form screen measured real but holds no production role. Ray Taulbot’s recency screen turned out to be already priced in, and three more of his angles, tested this week, came back unsupported or untestable. Tom Brohamer’s velocity engine failed calibration and is parked. Steven Davidowitz’s key-race concept could not be tested on our archive. Gary Hall’s two testable claims failed this week. Bill Benter’s published market-blend research informed our fusion study, which failed; the blend that ships is our own construction. Andrew Beyer’s figures served once as a cross-check, never as an input.
Books supplied the ideas. The record decided which ones stayed — and the record comes from public sources:
The permanent page: this piece also lives on the site as a reference document — The Books Behind the Method — linked from the homepage footer.
The picks, the pace sheets, and this podcast are all produced by AI. The system behind them was built and tested using many of today's leading AI systems in a strict deterministic environment set up as a handicapping lab, and it has been independently audited by other Large Language Models. Rankings follow the morning line.