📊 Findings & receipts

Every treble leg ever published here, judged against the Betfair Starting Price. A pick that consistently beats the close has real value whatever its result that day. A pick that loses to the close and wins anyway just got lucky.

All-time: 42 legs · average CLV +1.1% · median +3.4% · 62% of picks beat the close · last 30 days: avg +1.1% over 42 legs

A win-probability model flags every horse it believes the morning market underprices. Observe-only: it bets nothing, we just keep score in public, since 2026-06-07 (3 days so far). Settled at the flagged morning price and at SP. Small numbers grow nightly. It gets promoted to real picks only if this record earns it.

We tagged what changed for 441,246 runs against each horse's previous run — first-time headgear, class moves, trip changes, going switches, layoffs, new jockey, new yard — and scored every factor with A/E: actual wins divided by what the market expected. At bookie SP the average runner scores 0.78 (the margin), so 0.78 is par. Anything real must beat par in two separate time windows.

Verdict: not one structural change beats the market. The prettier the story, the more the market has already charged for it. The classic "new trainer will wake it up" angle is the worst of the lot — punters over-bet the romance.

Bookie each-way terms are a fixed formula, not a market. We tested the formula over 52,332 races against real results AND against the exchange's own place prices. The folklore says each-way is free money. It is not: at standard terms, no pocket is profitable on its own.

The one soft spot: big-field handicaps — 16+ runners, 4 places at quarter odds, horses priced roughly 9.0–29. There, and only there, the formula's place price beats the exchange's place price (+2% to +7%, both halves of the data, before results even enter it). The formula simply cannot price 4th place in a 20-runner cavalry charge.

The poison: 5–7 runner races. Two places at a quarter of the odds loses 20–40% on the place half in every odds band. The worst product in racing.

Everywhere else: a 6–15% structural drag any selection skill must climb first. Our each-way picker was rewired the day this was found: banned from small fields, held to a much higher bar outside the soft pocket.

Results lie for months — variance is that big. The closing price does not. If your picks consistently beat the price they close at, profit follows; if they consistently lose to the close, your winners are luck wearing a suit. That is closing line value (CLV), the fastest honest test in betting, and every pick this project makes — public or experimental — is now scored on it nightly against the Betfair Starting Price.

A real example from our own book: a pick taken at 11/4 in the morning drifted out to nearly 9.0 at the off — and won. The result said genius. The market said wrong price all day. Over hundreds of bets, only the market's verdict matters.