Introduction
Ryujin Radar's cumulative analyzed race count has passed 10,000 races.
Many AI prediction services exist, but very few publish their entire track record beyond 10,000 races. This article honestly shares what 10,000 races of data reveal about AI prediction's "strengths" and "limits."
Cumulative results summary
| Model | Races | Win hit rate | Win return rate | Trio hit rate | Trio return rate |
|---|---|---|---|---|---|
| Standard | 12,324 | 47.4% | 85.9% | 18.0% | 73.5% |
| Favorite-focused | 10,020 | 36.1% | 84.8% | 14.3% | 71.5% |
| Longshot | 10,020 | 25.2% | 81.4% | 9.5% | 69.2% |
These numbers are based on a total of 32,364 predictions.
AI prediction's "strengths" — the good news
1. What a 47.4% Win hit rate means
Boat racing has 6 boats per race. A random pick has roughly a 16.7% (1/6) chance of hitting.
Ryujin Radar's standard model hits 47.4%. That means:
- About 2.8x the accuracy of a random pick
- The favorite wins roughly 1 out of every 2 races
- Since this is measured across 12,324 races, it's not a fluke
2. Place hit rate is around 70%
The standard model's Place (top-2 finish) hit rate is about 70%. That works out to the favorite finishing top 2 in 7 out of every 10 races.
This is evidence that the AI's predictions aren't off base. Even when it misses 1st place, the pick often still lands in 2nd.
3. Trio hit rate of 18.0%, and Trifecta accuracy
A Trio bet requires picking the top 3 boats regardless of order. Its probability is 1/20 (5%).
Ryujin Radar's standard model hits 18.0% (about once every 5.5 races) — roughly 3.6x better than random.
| Model | Trio hit rate | vs. random |
|---|---|---|
| Standard | 18.0% | ~3.6x |
| Favorite-focused | 14.3% | ~2.9x |
| Longshot | 9.5% | ~1.9x |
Even for Trifecta, which requires the exact finishing order (probability 1/120), the standard model achieves 5.9% — about 7.1x better than random.
4. Broken down by venue, a return rate above 100% shows up often
While the overall monthly and average return rates fall below 100%, breaking things down by venue × model × bet type reveals many conditions that exceed a 100% return rate.
From December 2025's results:
- Ashiya × Standard × Trifecta = 1,258.9%
- Hamanako × Longshot × Trifecta = 515.0%
- Marugame × Longshot × Trifecta = 382.4%
Rather than betting evenly across all 24 venues, focusing on venues where the AI performs well matters.
AI prediction's "limits" — the honest part
1. Overall return rate is below 100%
| Model | Win return rate | Trio return rate |
|---|---|---|
| Standard | 85.9% | 73.5% |
| Favorite-focused | 84.8% | 71.5% |
| Longshot | 81.4% | 69.2% |
Every model's return rate, betting flatly across all races, falls between 69% and 86%.
What does that mean? Betting ¥100 flatly on every race would, over the long run, mean losing about 15% of your total wager.
2. Why the return rate falls below 100%
Boat racing's deduction rate is around 25% — meaning a ¥100 bet returns an average of only ¥75.
Ryujin Radar's 85.9% Win return rate clears "part of" that deduction wall, but doesn't fully break through it.
Random (expected value): ~75%
Ryujin Radar Standard: 85.9% ← +10.9% improvement
Ryujin Radar Favorite-focused: 84.8% ← +9.8% improvement
Break-even point: 100%
3. No such thing as an "AI that always wins"
This is a firm statement: no AI, applied across every single race, can reliably maintain a return rate above 100%.
Why:
- Boat racing has many real-time variables — start timing, wind direction, water conditions
- AI can only analyze the data available before the race starts
- Odds already reflect the collective knowledge of every market participant
So how should you actually use it?
Strategy 1: Focus on specific venues
Instead of betting across all 24 venues, concentrate on the venues where the AI performs best.
Ryujin Radar has custom analysis rules built for 7 venues, and accuracy tends to be higher at these venues.
Strategy 2: Filter by condition
Ryujin Radar's predictions include a "confidence" score. Narrowing down to high-confidence predictions improves both hit rate and return rate.
Strategy 3: Use different bet types strategically
- For stability: The standard model's Place bet (~70% hit rate)
- For return rate: The favorite-focused model's Win bet (91.1% return rate as of January 2026)
- For a big swing: The standard model's Trio (18.0% hit rate) or Trifecta (100%+ common at specific venues)
Strategy 4: Manage your bankroll strictly
AI prediction is ultimately a reference tool. Always follow these rules:
- Decide your daily budget in advance
- Don't chase your losses with additional bets
- Never dip into your living expenses
- Also check the Responsible Gambling page
Looking ahead
Continuous model improvement
Ryujin Radar analyzes new data every month to keep improving its models.
- Expanding venue-specific rules: Currently 7 venues → targeting all 24
- Factoring in seasonal effects: Analyzing the relationship between temperature, wind speed, and motor performance
- Leveraging real-time data: Exploring how to incorporate start-exhibition data
Toward 20,000 races
The next milestone is 20,000 races. As data accumulates, the AI's analysis accuracy improves. We'll keep publishing the full track record as we continue improving.
Data transparency
About the numbers in this report
- Every number matches the data published on Ryujin Radar's accuracy page
- Good and bad numbers alike are published without exception
- Calculated using a flat ¥100 bet per race (no bias in bet sizing)
"Can I really trust this?"
Ryujin Radar publishes its predictions in advance, and checks the results only after the fact. There's no after-the-fact editing whatsoever. Every prediction can be verified retroactively in the race history.
Summary
- Across over 10,000 races, Win hit rate is 47.4% (~2.8x random)
- Trio hit rate is 18.0% (~3.6x random), and Trifecta hits 5.9% (~7.1x random)
- Return rate falls below 100% at 85.9% for Win and 73.5% for Trio — betting flatly on every race doesn't win
- Narrowing to specific venues or conditions can easily push return rate above 100% (e.g., 1,258.9% for Ashiya Trifecta in December)
- AI prediction isn't infallible. Best used as a reference, combined with your own judgment.
Ryujin Radar builds trust by publishing everything.
Tags: #RyujinRadar #10000Races #AIPrediction #TrackRecordAnalysis #HitRate #ReturnRate #DataDisclosure #BoatRacing