Hong Kong racing has one of the richest public datasets in world sport — yet most punters still bet on gut feel. At Mystique Racing we built a fully quantitative pipeline and publish every prediction before the race, so our track record is auditable by anyone.
The model in one paragraph
We blend opening-odds baselines, live drift (steam moves), sectional-time features and jockey/trainer form into a single win-probability model. After every meeting, an EMA (exponential moving average) calibration loop nudges model weights toward whatever the market is currently rewarding — no manual overrides, no hindsight edits.
What "drift detection" catches
When a horse's odds shorten 20%+ from the opening line in the last hours before a race, that's information. Our guard flags it, measures whether the drift is steam (smart money) or noise, and adjusts the predicted win percentage accordingly. On a recent card, drift signals correctly identified the two strongest late moves of the night.
Radical transparency
- Every pick is timestamped pre-race and archived
- Calibration changes are logged in a public parameter history (24 adjustments and counting)
- We publish our misses. A model that can't show you its losing days is marketing, not science.
Full predictions, results and methodology: mystique-racing.com
Built in Hong Kong. Written by the pipeline, reviewed by humans.
Top comments (0)