What every label on the race card and results pages means, in plain English.
On the race card
Model win %
The model’s estimated chance for each runner. The percentages in a race add up to 100%.
Model price
That chance expressed as decimal odds (100 ÷ model win %). A 25% chance is a model price of 4.0.
Odds · Mkt
Live HKJC win odds (decimal, including your stake), and the runner’s rank in the market — 1 is the favourite.
Gap (pts)
Model % minus market %. Positive means the model rates the runner higher than the market does. It is information only, not a betting signal: in testing, backing horses the model rated above the market lost money.
Top Rated
The model’s highest-rated runner in the race.
Threat
The model’s second-highest-rated runner.
E/W top 3
The model’s three highest-rated runners — the ones it rates most likely to finish in the first three.
Runners to note
A short factual rundown of the top three: jockey, trainer, draw, last finish, days since that run, and record at the distance and course.
Race status
Early
Ratings from form before today’s results are in. Going is assumed until HKJC declares it.
Live
Ratings updated with live odds and today’s earlier results. Still able to change.
Rated (locked)
Ratings locked 4 minutes before the off. A locked rating is never changed.
Feed error
Race data was incomplete at the lock, so no rating was published for that race. This status is permanent.
Not locked
Our system was not running at the lock, so no rating was published for that race. Counted on the record, not hidden.
Off
The race has started.
Archived terms (Sunday V2 test, 27 Sep 2026)
WATCH
The leading runner before the lock — never a bet.
BET / FINAL BET
The runner that met the V2 rule at the lock (estimated value of at least 10%).
PASS / FINAL PASS
No runner met the rule, so no bet.
FEED_ERROR
Race data was incomplete at the lock, so no bet was made.
Performance figures
Strike rate
The share of selections that won.
P/L and units
Profit or loss. A unit is one level stake; V1 paper bets were 1 unit each.
ROI
Return on investment: total profit or loss divided by total staked. −10% means 10p lost per £1 staked.
95% range
On How we test, where a return would plausibly land given the luck involved (a bootstrap interval). For example, the V2 rule’s range was -47.2% to -3.2% — entirely below zero.
Log loss
A score for how much probability a method gave the actual winner. Lower is better.