About TBredIQ · the masthead

Independent. Data-anchored. Intelligence-only.

We publish a daily read of thoroughbred racing in two layers: plain-English up top, the full data underneath. We don’t take wagers, recommend stake sizes, or tell you how to play. The analysis is the product.

We don’t tell you what to bet. We tell you what the data says.

Who we are

An analytical publication for thoroughbred racing.

Same data the professional handicapping room stares at — past performances, connections, pace, class, surface, weather, market behavior — assembled into composite signals and published every race day. Two reads of every race: a plain-English brief for the casual visitor, the full data layer for the sharp.

What we’re not: a sportsbook, an ADW, a tipster service, or a bet facilitator. We don’t accept wagers. Everything we publish is analytical context you can take to your licensed wagering operator of choice — or just to follow along on a Saturday.

Editorial scope

What we publish. What we won’t.

The line between intelligence and enablement matters. We sit on one side of it on purpose.

We publish
  • Calibrated model probability and edge per runner
  • Race-shape classification — standout, mid, open
  • Per-track priors · bias, pace, broken-track flags
  • Closing-line value once capture coverage is sufficient
  • Carryover snapshots and pari-mutuel split-pool math
  • Plain-English analyst notes generated from the same data
We don’t
  • Tell you a stake size or ticket structure
  • Track your wagering activity or P&L
  • Accept wagers or facilitate betting in any form
  • Run affiliate links or referral fees from any operator
  • Use phantom-bet ROI projections to drive engagement
  • Pad the card to fill space — some days fire zero
How we measure

Three honest tests, published openly.

Most retail handicapping products quote “+30% ROI” without naming the bet structure, the takeout, or the sample. We publish methodology before we publish numbers.

Test № 01

Closing-line value

Did the picks beat the market’s closing line, or were we lucky? CLV is the gold-standard sharp-vs-square test. Capture runs every 15 minutes during racing windows.

Test № 02

Calibration

When the model says X% conviction, does the bin actually win X% of the time? We publish the reliability table at performance with drift per bucket.

Test № 03

Race-shape gating

Some races are wide open by the model’s own measure — we mark those down. Standout shape gets promoted. We never trumpet a flat race.

Data layer

Where the numbers come from.

Primary feed: theracingapi.com Standard tier — full North American, UK, and Irish racecards, results, payoffs, runner-level form, comments, equipment markers. Brisnet drops ingested manually for past-performance archive depth.

№ 01

Trainer / jockey 30-day stats

Computed from rolling /results aggregations, broken out by surface where the sample size allows. Top-tier thresholds are recomputed every Sunday.

№ 02

Track bias signals

Last-14-day post-position win rates, flagged when there’s a 10pp+ lift over the uniform baseline. Five tracks are explicitly marked as broken — historically negative-ROI venues we treat with caution.

№ 03

Carryover history

Daily snapshot of every active rollover with day-over-day delta. Pool dynamics and split-pool math live on /carryovers — informational, no “play this” verdict.

№ 04

Trip notes & equipment

Thirteen-keyword taxonomy parsed from comment + spotlight fields (checked, stopped, wide, rallied, etc.). First-time-with markers on headgear, wind surgery, days since last run.

№ 05

Sectional times

Quarter / half / three-quarter / final extracted from /results fraction objects. Pace profiles and shape priors derive from this layer.

The model

A LambdaRank ranker, three filters, no magic.

The pick selection model is a LightGBM LambdaRank trained on multi-year North American data. Inputs: morning-line, post position, distance, surface, recent form figures, trainer + jockey rolling stats, sire-distance fit, class. Output: a per-runner ranking that orders the field by predicted finish.

A runner only surfaces on the published card when the qualified-plays filter also fires — top-tier trainer and jockey on a rolling 30-day window, fair-value odds (morning line 4–1 to 12–1), and a race-shape that’s not flat. Public favorites pay nothing on the signal; the band is where the math actually works.

The 12–15% WIN ROI in the 90-day backtest is descriptive of past performance, not predictive — and we say that everywhere it appears. — Methodology, /performance

Posture

Intelligence, not enablement.

Several US states regulate “tout services” and similar products. We deliberately stay on the analysis side of the line so that TBredIQ remains usable everywhere thoroughbreds run. Internally, every feature passes through a check: is this showing data or telling the user how to bet? Anything in the second bucket gets reframed before ship.

  • The pari-mutuel calculator displays pool dynamics and split-pool math — no “play this” verdict.
  • The race-day timeline shows the analytical card — no recommended ticket structure.
  • Spotlight and Sharpest Spot describe analytical setups, not stake sizes.
  • Yesterday’s replay shows finish positions — never phantom P&L on bets you didn’t actually make.
Compliance

Independent & transparent.

  • Every footer carries the “informational only — TBredIQ does not accept wagers” disclosure.
  • No affiliate referral fees from any racetrack, ADW, or wagering operator.
  • localStorage usage (theme, watchlist, dismissed onboarding) disclosed via the cookie banner.
  • No third-party tracking pixels, analytics scripts, or ad networks.
  • Backtest and calibration numbers recomputed weekly and published to /performance.
Stack

What it’s built on.

  • Backend · Python / Flask · SQLite (WAL) · Anthropic Claude for editorial AI.
  • Frontend · vanilla HTML + CSS + ES6 (no framework). Self-hosted fonts.
  • Mobile · PWA + iOS / Android Capacitor wrapper.
  • Cron · Railway scheduled jobs orchestrate the daily pipeline + 15-minute live windows.
  • Tests · 413 passing across backend, email, ops scripts.
Independent

Funded by direct subscriptions.

TBredIQ is not affiliated with any racetrack, breed registry, past-performance vendor, ADW, sportsbook, exchange, or bet processor. We don’t accept advertising. We don’t resell user data. The product is funded by direct subscriptions and one-time pass purchases. That alignment is intentional.

Questions about methodology? support@tbrediq.com