Hong Kong Racing Research Agent
An agent that reads only what the Jockey Club publishes: racecards and draws, trackwork, horse and jockey history, and how the market's own prices move. Probabilities are sealed before the gates open and checked race by race against the official result afterwards.
Research engine running · Telegram agent open
What it does today
@Oddsflowjockey_bot is open on Telegram: the next meeting’s card, the latest results and the reading guide are all one message away. It leads in Traditional Chinese and also speaks Simplified Chinese and English. Model signals follow once the upstream endpoints are ready.
What it has read so far
- race meetings
- 181
- races
- 1,763
- horses
- 1,985
- jockeys and trainers
- 92
- trackwork records
- 1,108,471
As of 21 September 2026, from the public OddsFlow Racing pages.
What it reads
Racecards and draws
Class, distance, surface, course and barrier for every declared runner, from the moment the card is published.
Trackwork
Over a million morning gallop records — the part of the form nobody reads in full, and the part that changes most between meetings.
Horse and stable history
Past runs, jockey and trainer pairings, recent condition, and how a horse has gone over this course and distance before.
How the market moves
Price movement across the day, and which runners move far enough to be worth a second look. The market is read as a signal, never followed as instruction.
The courses it reads
Sha Tin · Turf
The larger of the two courses. Starts from 1000m to 2400m sit around the whole circuit, with a long straight and comparatively little bend.
Happy Valley · Turf
A 1,450m circuit, so almost every race is run round a bend. Most of the argument about barrier draws happens here.
Sha Tin · All-weather
A tighter circuit on an all-weather surface. The same distance runs more bend here, and it does not ride like the turf.
Conghua · Turf
The mainland training and racing site. A different configuration again from Sha Tin and Happy Valley, and the thinnest history on file.
Course diagrams from OddsFlow Racing, marking where each distance starts. The same distance runs a different amount of bend on a different course — which is why the barrier draw is worth analysing on its own.
How it is held to account
Sealed before the gates, with a fingerprint
Probabilities are frozen before the first race closes, and the seal time and a SHA-256 digest are published with them. A digest that no longer matches means the contents changed — a lock anyone can check for themselves.
Settled on the official result, ledger public
Each leg is decided at the live price on the clock at cutoff, settled against the Jockey Club’s published result, and written into the ledger line by line. The stake is virtual throughout; no money is placed.
Two series, kept apart
The registered forward window and the two earlier runs of an unconverged model are recorded and shown separately. Merging them would make the numbers look better, which is the reason they are not merged.
Losses stay in the record
A race the model got wrong stays exactly where it is. A record you can only read after it has been filtered is not a record.
A record, not a tip service
This agent publishes probabilities and the way they were reached. It makes no decisions for you and takes no position of any kind.
How it relates to the football agent
Same protocol, separate interface. The football agent and this one run the same loop — read the data, publish before the event, settle against the real result, keep the whole record — on two sports that share nothing else. Neither one's record is mixed into the other's.
See the football recordOpen now
@Oddsflowjockey_bot is running on Telegram. Ask it for meetings, results, horses and jockeys — no sign-up.
Open the racing agentOddsFlow Racing is a research project. It is not affiliated with, endorsed by, or connected to the Hong Kong Jockey Club, and it uses only information the Club publishes.
The next meeting, sealed
2026-09-23 · Happy Valley9 races, 108 declared runners. Below is the model’s win probability for each — within a race they sum to exactly 100%. Tap any race for the full field.
Race 111. 東方魅影39.2%
- 11東方魅影39.2%
- 3神駒馬靈20.5%
- 8電訊驕陽8.6%
- 4紅磚戰士8.1%
- 10威威父子5.9%
- 6開心三多4.5%
- 9至高心得3.9%
- 12幸運同行3.6%
- 1堅多福2.4%
- 5極速滿貫1.4%
- 2一風雲1.2%
- 7綫路達飛0.7%
Race 24. 馬馳登18.1%
- 4馬馳登18.1%
- 3路路勁13.2%
- 7銀刺勇士12.0%
- 10比特星11.9%
- 12首駿11.5%
- 9開心五月11.2%
- 8鄉村威龍10.1%
- 1紅愛舍4.3%
- 5飛輪霸4.2%
- 11有盈勇士2.5%
- 2銳喜0.6%
- 6金珀尖子0.5%
Race 35. 凝妙星18.0%
- 5凝妙星18.0%
- 2贏玥15.4%
- 8蹺妙11.6%
- 6領航天子9.7%
- 3彩虹七色9.2%
- 7有情有義9.1%
- 9越駿齊歡6.8%
- 4睿智多寶5.9%
- 1星願無限4.1%
- 10家傳之寶4.1%
- 11光明傳承3.1%
- 12得意佳作2.9%
Race 46. 快樂神駒29.0%
- 6快樂神駒29.0%
- 3應龍飛影18.0%
- 9將傲16.0%
- 4星辰千帥10.6%
- 8禪勝閃亮8.6%
- 11焦點5.9%
- 5莊家班5.0%
- 1沙井之友2.1%
- 2智勝一籌1.8%
- 7震撼人心1.1%
- 12三強1.0%
- 10平天雄0.9%
Race 56. 越駿聯歡17.9%
- 6越駿聯歡17.9%
- 9爆竹14.0%
- 7滿洛城10.9%
- 3大學生10.5%
- 1本領非凡10.1%
- 5創科群英9.3%
- 4赤風驪7.3%
- 12同心5.1%
- 8金駒永騰4.7%
- 11健康小馬3.8%
- 2大文豪3.7%
- 10揀馬之皇2.7%
Race 64. 巴閉王26.2%
- 4巴閉王26.2%
- 1福進20.2%
- 12馬運高14.5%
- 5佐治傳奇9.8%
- 9天火同人9.7%
- 11萬眾開心6.8%
- 6升升雙息3.6%
- 2友駿同心2.7%
- 7雙劍合璧2.6%
- 3藍地球2.2%
- 8螢影飛馳1.0%
- 10領航多財0.8%
Race 78. 首飾悟空23.6%
- 8首飾悟空23.6%
- 3天星11.3%
- 2乘數表9.6%
- 4競駿皇者9.3%
- 6電源之駒9.0%
- 1東來欣賞7.3%
- 11盈妍威楓6.9%
- 7加州本事5.7%
- 10正極5.3%
- 12丞匡掠影4.7%
- 9安康萬里4.3%
- 5傲聖2.9%
Race 811. 驕陽雄心21.4%
- 11驕陽雄心21.4%
- 8繼往開來16.3%
- 6富國兄弟10.9%
- 10喵喵怪9.8%
- 3富心星9.3%
- 1人和家興7.8%
- 4久久為昇5.3%
- 2團結勇士5.2%
- 5飛馬座4.4%
- 7勇霸龍4.2%
- 9蓮冠皇4.0%
- 12亞機拉1.4%
Race 92. 紫荊傳令21.0%
- 2紫荊傳令21.0%
- 1嘉應傳承12.7%
- 12財富非凡10.7%
- 3凌登9.3%
- 8豐辰7.8%
- 6將義7.7%
- 5好實力7.2%
- 7浪漫鬥士7.2%
- 10風將6.5%
- 9中國心4.4%
- 4平凡騎士2.8%
- 11瑪瑙2.7%
Model v1F 2026-09-01 / v2 2026-09-01 · racecard 2026-09-21. Probabilities are sealed before the gates open and checked against the official result afterwards. Horse names as the Jockey Club publishes them.