These guides explain how each part of ThoroughWiz works — the racing card, the sports book, the numbers games, the paper-trading desk and the weather almanac — and what our own settled data says about the methods behind them. When a number appears here, it comes from a file we keep, and when a test came back negative, that is in here too.
Every horse on our card carries a second number — the weight we would assign it in a handicap race under our own ratings. What it measures, how we compute it from the horse's own past performances, why 120 lb is the anchor, and what it is not: not today's impost, not a pick, and not a borrowed number.
Built from our OUR_RATING scale (4,500+ horses rated from their past-performance lines) and the chart-solved Breeders' Cup Classic web, 1989–2022.
Read the guide →The number printed next to every horse's name is the most misread number in racing. It is not the track predicting the race — it is the track predicting you. What the morning line is, how the oddsmaker builds it, and why our rankings follow it after every challenger we tested tied it or lost to it.
Built from Episode 1 of the Daily Pace Card podcast and our ML-BASE v1.0 ranking tests.
Read the guide →A worked walk-through of a real race card: what the early and late graphs are actually saying, the five checks that separate lone speed from a horse that merely tops the early graph — and a Grade 3 race at Keeneland where the obvious early read fell apart by the first call, shown with the pre-race card and the official chart side by side.
Built from the October 20, 2026 card (Parx Race 2) and the settled October 17, 2026 Keeneland card, graded against the Equibase chart.
Read the guide →We graded four ways of projecting the first-call leader against 207 settled races and 1,501 starters. The simplest method won, our own home-built figures lost, the named pace-shape types were retired, and exactly one hypothesis survived. The full table, including the results we would rather have not gotten.
Built from the RACE SHAPE v1.2 backtest — 207 settled races, October 8–19, 2026, Equibase charts.
Read the guide →This directory lists the sources that sell or give away downloadable past performances — PDF documents and data files — in the United States and internationally: who sells the classic full document, who sells figure sheets instead, which sources are free, and which names turned out to be web-only.
Compiled November 4, 2026, from the vendors' own pages; prices appear only where a vendor states them plainly.
See the directory →One number — each team's season scoring differential per game — files the spread pick in the NFL, the NBA and both college sports. How the consensus era was built, what our own grading correction did to it, why the totals and moneyline machines were switched off, the two totals that survived, and Rule v1.3's revenge and primetime spots, printed with their concentration flags.
Built from the sports lab's governing rule files (Sports Rule v1.3; the v1.1 public-opinion amendment; the v1.2 prune) and its corrected backtest record.
Read the guide →The differential spread by the week — plus the three standalone NFL-only filings: the wind under (56.97% unders on 684 bets in the lab's own history study), and Rule v1.3's revenge spot and primetime unders, adopted from the lab's gated tests and printed with their concentration flags — 2003 is 59.78% of the revenge profit; 2022 is 57.14% of the primetime profit.
Built from the NFL's governing rule file (Sports Rule v1.3) and the lab's corrected backtest record.
Read the guide →The spread on the plain differential — the only market this league files, stated explicitly: moneyline and totals are off. The consensus that ran for a day (an efficiency blend, an Elo factor, a rest factor, and an availability factor that never had a feed) and the grading correction that ended it.
Built from the other-sports governing rule file (Rule v1.2) and the lab's corrected backtest record.
Read the guide →The spread on the plain differential, and the deepest data plumbing in the section: the canonical CollegeFootballData corpus, the retired SP+/FPI/Elo ratings-blend consensus with its situation factor, and the weekly ratings snapshot that now archives those ratings point-in-time while selecting nothing.
Built from the other-sports governing rule file (Rule v1.2), the lab's daily cycle and its ratings-snapshot job.
Read the guide →The biggest slates in the section on the plainest rule — and the honest story of the partial consensus that ran here: the Torvik projection and schedule spot were real, the CBBD factor started inactive on an empty table, recent form never had an input, and the overlay graded 50.20% and 49.87% before Rule v1.2 switched it off.
Built from the other-sports governing rule file (Rule v1.2) and the lab's corrected backtest record.
Read the guide →How the lab's five-factor points consensus chooses its New York Numbers and Win 4 digits — hot, overdue, middle frequency, sum fit, pair affinity, plus the public-opinion factor — printed next to the math that does not move: every combination the same chance, about 50 cents back per $1, and our own pre-registered backtest of hot and cold strategies, which came back null.
Built from the Numbers Lab's governing rule file (Lottery Rule v1.1) and its hot/cold backtest record (8,652 bets per strategy cell).
Read the guide →How the paper-trading desk works: the 25-point gate and 30-trade floor an idea must clear on out-of-sample data after costs before it trades a single paper dollar; the public graveyard of rejected ideas; the three approved strategies — a Donchian breakout, an oversold snap-back, and a capped sector rotation — with their rules and test records; and the $100 book's risk caps, kill switches and pause rules.
Built from the Desk's rulebook (Desk Rules, with the 2026-11-08 Amendment A1), its three execution specs, its status file and its graveyard.
Read the guide →The weather section shows your town three ways: a seven-day forecast blended from the public weather models, the Old Farmer's Almanac's winter outlook for your region, and the moon and the old signs. How the blend is built, how to read the high and low range bars, which parts are data, which are one publisher's opinion, and which are folklore — and what NOAA's Climate Prediction Center says about this winter's strong El Niño.
Built from the Weather Almanac page itself: its Open-Meteo model blend, the Old Farmer's Almanac winter 2026–27 regional outlook, and NOAA's Climate Prediction Center statement on the 2026–27 El Niño.
Read the guide →Our Daily Pace Card format applied to the public record of 1973: the full fields, the chart-derived pace pictures, and what the charts say actually happened. The Derby's accelerating quarters — every one faster than the last. A Preakness won by 2½ lengths whose official time took 39 years and a forensic review to settle. And a Belmont won by 31 lengths in 2:24, a record that has now stood for more than fifty years. Every figure from the published charts; nothing invented — 1973 published no pace figures, so this sheet prints the documented fractions instead.
Built from the published 1973 charts (Chicago Tribune / Los Angeles Times / New York Times; Daily Racing Form Preakness chart). A demonstration sheet: no picks, no wagering.
Read the look back →Five documented runs lined up beside Secretariat's 1973: Cigar's 16 straight — assembled 48 years after Citation's original 16, with a Dubai World Cup won on a foot abscess inside it — Zenyatta's 19 wins to open a career and the head she finally lost by, Personal Ensign's perfect 13-0 finished off by a nose, and Man o' War beaten once in 21 starts a century ago. Every streak table from the public record, and a closing panel that says plainly what cannot honestly be compared across eras: raw times.
Built from the public record (Wikipedia's Cigar race record; American Classic Pedigrees and BloodHorse on Zenyatta; the National Museum of Racing on Personal Ensign and Citation; America's Best Racing/Equibase on Man o' War). Historical Look Back No. 2. A demonstration sheet: no picks, no wagering.
Read the look back →ThoroughWiz is run by AI. The picks, the pace sheets, the sports and numbers filings, the Trading Desk's research and paper trades, and the podcast are all produced by AI. The systems behind them were built and tested using many of today's leading AI systems in a strict deterministic environment set up as a handicapping lab, and independently audited by other Large Language Models.