賽馬賠率是某一刻的市場價格，不是真實勝率 Horse-racing odds are market prices at a moment, not true win rates 赛马赔率是某一刻的市场价格，不是真实胜率

A price of 6.3 is a return quote, not a 6.3% chance

HKJC's displayed odds include the returned stake. Its official example explains that odds of 11 return HK$110 on a winning HK$10 ticket: HK$100 profit plus the HK$10 stake. On the same convention, odds of 6.3 return about HK$63 on HK$10 if the runner wins.

Taking the reciprocal gives a useful raw implied probability: 1 ÷ 6.3 = 15.87%. The word raw matters. The figure has not accounted for the prices of the other runners or the pool deduction, so it cannot be labelled the horse's true win rate.

Why did the full field add to 123.04%?

In a sellable snapshot of Happy Valley Race 1 at 18:29:57 on 15 July 2026, the reciprocals of the 12 WIN odds totalled 123.04%. Treating each reciprocal as a final probability would allocate 123.04% of winning chance to one race, which is not a coherent probability distribution.

The excess over 100% reflects the structure and display of pool prices. It shows why one runner's reciprocal is not enough to understand how the market allocated relative chance across the field.

Snapshot: 2026-07-15 18:29:57.118 Hong Kong time. The pool was still selling; these are not official final odds.

Normalization fixes the total, not the market's judgment

Normalization divides each raw reciprocal by the field total. For LUCKY BLESSING, 15.87% ÷ 123.04% = 12.90%. The adjusted field now totals 100%, giving a consistent description of the market's allocation of relative chance at that moment.

The 12.90% remains a market probability, not a true win rate established by nature. Going, pace, horse condition, limits in public information and collective judgment can all separate price from outcome. Normalization solves an adding-up problem; it does not solve prediction accuracy.

Across 1,712 races, 22.73 points is not 17.5%

The archive contains 21,123 runner records across 1,712 races from 10 July 2024 to 8 July 2026. The mean reciprocal total was 122.73%, the median 122.68%, and the range 103.39%–128.25%. All 1,712 totals exceeded 100%, showing that raw reciprocals normally do not form a ready-to-use probability distribution.

The 22.73-point excess over 100% is not the same quantity as the 17.5% pool deduction. HKJC states that 82.5% of the WIN pool is paid as dividends; the simplified reciprocal is 1 ÷ 0.825, or about 121.21%. Price rounding, snapshot timing, pool state and stored precision may contribute to the observed 122.73%, so the figures should be reported separately.

The archive summary describes stored market snapshots; it is not a fixed takeout formula.

The same runner moved from 6.3 to 4.0 because time changed

LUCKY BLESSING was 6.3 in the sellable 18:29:57 snapshot; the official result page later listed final WIN odds of 4.0. Both numbers can be valid because they come from different moments. As money enters a pari-mutuel pool and each runner's share changes, the displayed odds update.

Turning odds into analysis requires three steps: preserve the timestamp, normalize within the field, and compare with an independent estimate from the same time. An early quote, a late sellable snapshot and final odds cannot be treated as one interchangeable market observation.