Backtest Simulation Result

AI vs Random
Benchmark Results

We ran a empirical backtest simulation using historical data to verify if our LLM (AI) models could statistically outperform pure random selection with special prompt and statistical analysis.
(Updated Jan 24, 2026)

10.59%
2+ Match Rate (AI)
+170% vs Random
99%
Empirical Confidence (Backtest)
P-VALUE: 0.0099*
1,350
Total Comparisons
135 Gens x 10 Draws

*Calculated based on 1,350 comparisons in a simulated environment. Results may vary in future draws; this is not a guarantee of success.

Model Performance Leaderboard

Comparing top Gemini Models against random baseline

RankModel NameTotal ScoreEfficiencyAttempts
1
Gemini 2.5 Flash
Jackpot Hunter
6,030 pts502.5 12
2
Gemini 3 Flash (Preview)
New Entrant
5,160 pts198.5 26
3
Gemini 2.5 Pro
Highest Efficiency
5,110 pts212.9 24
4
Gemini 2.0 Flash Lite
Most Reliable
1,380 pts23.8 58
5
Others (Mixed)
Claude 3.5, GPT-5 Nano, etc.
1,150 pts76.7 15
Scoring System:
5 Match: 10,000pts4 Match: 5,000pts3 Match: 1,000pts

Powerball Real-World Simulation

Methodology: To ensure statistical rigor, we conducted a massive backtest simulation. We took 135 unique AI-generated sets and cross-referenced each one against the latest 10 official Powerball draws (2025.12.29 - 2026.01.19). This created 1,350 distinct data points, which we then compared against an equal number of random sets to isolate the AI's predictive edge.

Matching ResultsTheoreticalRandom (1,350)AI Model (1,350)The Edge
0 Numbers (Total Loss)67.84%68.76%50.74%-18.02% Loss
1 Number Matched28.27%27.23%38.67%+11.44% Acc
2 Numbers Matched3.71%3.92%9.56%2.44x Better
3 Numbers Matched0.18%0.077%0.96%12.5x Dominance
4 Numbers Matched0.002%0.00%0.07%AI Exclusive
Efficiency Insight

Probability Density

Matching 4 numbers in Powerball mathematically requires 36,525 tickets on average.
Our AI achieved this result in just 135 generations. This suggests a potential Search Space Optimization. It implies that the AI model may be identifying favorable patterns more frequently than random chance, potentially increasing the 'density' of winning numbers within a smaller sample size.

Efficiency Multiplier27x
Sample Size135 Sets
Random Luck (Expected)36,525 Tries
AI Model (Actual)135 Tries

Target hit 27x faster

Transparent Insight: The Bonus Ball Gap

While our AI excels at Main Number Combinations (beating random by 2.5x), the Powerball (Bonus) prediction remains as volatile as random selection. We believe in showing the raw truth of data.

Random Factor: 100%
AI Logic: Main 5 Numbers

Hall of Fame

Backtest Results
4
Highest AI Achievement

4 Numbers Matched

Draw Date: 2025.12.29 (Gemini 2.5 Flash)

$100.00
Prize Value
3
3 Matches
2026.01.19
$7
3
3 Matches
2026.01.14
$7
2+
2 Matches + PB
2026.01.10
$7
+ 16 more winning sets found

Common Questions

Understanding the data and limitations

LottoKim AI - AI Lottery Generator that explains the reasoning behind | Product Hunt