Guide & key

Predictions come from a deterministic blend of three signals โ€” model 35% (Claude's 12-factor score), market 45% (live fixed odds) and VectorOdds 20% (our own composite AI rating). The flags below surface where the model, the market and the tote disagree.

Alerts page

๐Ÿ’ฐ VALUE +18%
Value A backed selection (top pick or each-way) is available at fixed odds meaningfully above its model fair price โ€” at least 15% over, capped at +80% (beyond that the fair price is treated as stale). An early bet may be profitable (+EV).
๐ŸŽฏ PLACE VALUE
Place value The same place-floor signal as on the race card, surfaced across the schedule: a strong model place chance (โ‰ฅ45%) the tote place pool is underbetting, with early speed confirming the floor. Shows near the off (needs the tote pool). A good each-way / exotic anchor.
โŸณ PICK MOVED
Pick moved The live market has moved enough that re-blending on current odds changes the race's top pick to a different horse. The alert shows the old โ†’ new pick and why.
โšก OVERLAY ร—1.7
Tote overlay The tote is paying meaningfully more than the fixed price on a competitive runner (ratio โ‰ฅ 1.30) โ€” value to back on the tote. Only shown within 30 min of the jump, once the tote pool has firmed up.
โš  DRIFT TRAP
Drift trap (dummy drift) The fixed price is drifting out (โ‰ฅ25% over 30 min) while the tote stays short (โ‰ค85% of fixed) โ€” a warning the smart money may be against a horse whose fixed price looks tempting.

Race page โ€” runner badges

WIN
Win The model's top pick โ€” its single best-value win selection for the race.
EACH-WAY
Each-way A place-value runner worth backing each-way (win + place).
AVOID
Avoid The model rates this runner one to leave out โ€” it's never tipped as the win or each-way selection, even if it ranks well on a single factor. It's a holistic call from the 12 factors; the runner's one-line reason on the card gives the specific "why". The common triggers are:
  • Form against it โ€” poor or declining recent figures, well beaten last start with no excuse.
  • Unsuited today โ€” unproven or poor record at this distance or on today's going (e.g. no wet-track form on a soft surface).
  • Class / weight โ€” stepping up sharply in grade, or carrying a big weight in a handicap.
  • Fitness query โ€” long layoff, or first-up with a weak first-up record.
  • Wrong end of the run โ€” weak speed / sectional ratings (no early speed and no closing burst), especially a backmarker in a race likely to suit on-pace types.
  • Draw โ€” an unfavourable barrier for the track and distance.
  • Connections out of form โ€” poor jockey/trainer strike rate at the track, or a stable going quietly.
  • Trouble / temperament โ€” stewards notes flagging it hung, raced greenly or began awkwardly, without a genuine hard-luck excuse.
An Avoid isn't a prediction it will run last โ€” just that it's poor value to back. If the live market later drifts it further, the โš  drifted flag reinforces the read.
VALUE
Value tag Shown on the fair-odds row when the live market price sits modestly above the model's fair odds (ร—1.15 to ร—1.8) โ€” a genuine overlay rather than a stale/drifting price.
โšก OVERLAY ร—1.7
Overlay Same tote-over-fixed signal as on the alerts page, shown against the runner near the off.
โš  DRIFT TRAP
Drift trap Dummy-drift warning against the runner (see the alerts entry above).
โš  drifted ร—2.1 ยท pick may be stale
Drifted The live TAB fixed price has blown out โ‰ฅ1.6ร— from its opening price โ€” the market has turned against this runner, so its prediction and rank may be stale.
๐ŸŽฏ PLACE VALUE ยท LEADER
Place value The model rates a strong place chance (โ‰ฅ45%) that the tote place pool is underbetting (place dividend implies โ‰ฅ25% less than the model), and TAB's early-speed rating confirms the place floor (not a backmarker). A strong each-way / exotic anchor โ€” backing it can blow out your exotic dividends. Shows near the off, once the tote pool is reliable.
โšก speed: LEADER
Early speed TAB's early-speed rating band (LEADER / MIDFIELD / BACKMARKER), derived from sectional data โ€” how prominent the runner is likely to be early. Leaders and on-pace types have a higher place floor.
was top pick
Was top pick This horse was the top pick when the prediction was made, but the live re-blend has since demoted it (usually because it drifted).
live
Live This runner's rank, win% and fair odds were recomputed on current market odds (the model and VectorOdds scores are unchanged โ€” only the market moved).
SCR
Scratched The runner has been withdrawn and takes no part.

Race page โ€” banners

๐Ÿ’ฐโšก
Betting opportunities banner Appears above the runners to surface the strongest value overlays (model pick above fair) and tote overlays for the race in one place.
โŸณ
Top pick updated banner Shown when the live re-blend has moved the top pick โ€” e.g. Horse A โ†’ Horse B โ€” with the price moves that drove it. The original pick is kept struck-through so the change is clear.

Prediction & odds fields

Fair $
Fair odds The model's vig-free price = 1 รท blended win probability. What the horse "should" pay if the blend is right. Compare to the market price to judge value.
Win %
Win probability The blended chance this horse wins (model + market + VectorOdds).
Place %
Place probability Chance of finishing in the paid places (derived from the win probabilities across the field).
Model
Model score Claude's pure 12-factor rating (form, class, jockey, barrier, sectionals, etc.), before any market blending.
VO
VO rank / score The VectorOdds composite rating โ€” a strong, market-independent signal that doesn't drift with tote money.
Market $ ยท tote ยท pl
Live market odds Fixed win price, tote win (shown only within ~30 min of the off, once the pool is reliable) and fixed place. This is the market feed used everywhere in the app.
๐ŸŒ NZL
International A non-Australian meeting (the code shows the country/region).
Stewards
Stewards comments Trouble-in-running notes from a horse's recent starts (e.g. "held up", "checked") โ€” excuses that suggest it ran better than its finishing position.

Thresholds are tuning knobs and may change as more results accumulate. Small samples are noisy โ€” treat every flag as a prompt to look closer, not a guarantee.