Sly Man’s Bet of the Day

Method

The number that kept pulling me back today was +705.1% — the model's estimated edge on a horse priced 20/1 by the books. That kind of gap between implied probability and assessed win chance does not appear without a reason, and finding that reason is the whole job. The engine scanned today's card against a strict filter config: odds band 2.0–50.0 decimal (no odds-on, nothing longer than 49/1), field size 6–14 runners (cutting out the small noisy affairs and the big-field lottery races), and a feature weighting that places market movement at the top of the pile at 1.6, speed second at 1.5, and form third at 1.2. Trainer strike rate and going each carry 0.8, class change 1.0, and draw sits unbuilt at 0.5 — that last layer is reserved for v2 and contributes nothing today.

Today's universe

The day opened with 54 races declared and 555 runners. The odds filter trimmed that to 503 — removing all odds-on runners and anything beyond 49/1. Field size brought it down to 403, and a final pass dropping runners that collide with the lay-of-day list removed four more, leaving 399 runners to be scored. That is the pool from which the shortlist was pulled.

The shortlist

The top five by composite are Ceo Stealth Mode at 7.57/10, Rocket Boy at 7.30/10, Geometry at 7.20/10, Reformation at 7.19/10, and The Monkey King at 7.13/10. Rocket Boy is the most complete profile in the shortlist — speed, form, trainer strike rate, and a market move all scoring positively — and at 13/8 it is genuinely competitive. Reformation and Geometry are both Newcastle AW runners with strong steamers behind them. But the feature that tipped the balance for Ceo Stealth Mode over all of them was the sheer scale of the market_move signal at 20/1 — a +2.00 score on the highest-weighted feature in the model, combined with a form reading that is the joint-highest on the entire shortlist.

Why this horse

The read I keep coming back to is this: the market_move feature scores +2.00 — the maximum the model awards — meaning a strict majority of bookmakers shortened Ceo Stealth Mode and none drifted. That is a clean, unambiguous signal. When that happens at 20/1 in a Class 4 novice at Newcastle AW, somebody knows something. The weight on market_move is 1.6, the highest in the entire config, and this horse maxed it out. That is where the composite of 7.57/10 is built — not despite the price, but in part because of how loud the money is at that price. The form feature adds +1.40 to the composite, driven by a recent winning run coded as '1-' — the latest position is a first, and the decay-weighted average of +2.00 is the joint-highest form reading across the shortlist alongside Reformation. The speed feature is marginally below the field mean — SR 128 against a field mean of 130.2, a z-score of -0.11 — so I am not backing a speed standout here, and I will not pretend otherwise. That is the one feature pulling the score down rather than up. The case rests on the money and the form, not the raw speed number.

What worries me

Several features are simply unavailable for Ceo Stealth Mode: class change has no prior class to compare against, trainer strike rate falls below the noise threshold of five runners in 30 days, going data shows no career runs on good or adjacent ground, and draw bias is unbuilt. That is a lot of dark corners for one selection. The speed reading is marginally negative. Reformation runs in the same race at 9/4 with a stronger speed figure (SR 154 vs 128), a going record of 1-from-1 on good, and an equivalent form score — so the race rival is real and the favourite option exists if the reader wants it at a shorter price. The engine conviction is MEDIUM, and that sits correctly here: the market signal is loud, but the thin evidence base on going and the missing trainer data mean this is a one-strong-feature selection at a price, not a rounded case.

The bet

Ceo Stealth Mode · 20/1 · 1 unit · MEDIUM