· Real-Road Backtest

v2.8.12 · 5000-Shoe Real-Road High-Yield Backtest Report max_mult=4

Backtest time: 2026-06-10 · Algorithm: 5-algorithm ensemble + Reverse Martingale stake · Data source: evo_data.csv (313,259 hands of real road data)

Key Metrics

Net P&L
+56,960 ¥
From 10K → 67K

Return
+569.60%
6.70× bankroll

Total Hands Bet

68,635

W 34,488 / L 34,147

Win Rate

50.25%

vs 52.6% break-even

Max Drawdown

64.84%

@ shoe 3,531 · 87,600 ¥

Peak Bankroll
135,100
Pre-drawdown peak

Circuit Breakers

4,947

3-loss triggers, 5-hand pause

Bankruptcies
✓ None
0 across full 5000 shoes

Core Conclusion

5-Algorithm Ensemble + Reverse Martingale Stake Wins

On 5,000 shoes of real-road backtest, Reverse Martingale stake (max_mult=4) + 5-algorithm ensemble achieved Net P&L +56,960 ¥ (+569.60%), max drawdown 64.84%, 4,947 circuit-breakers, 0 bankruptcies. System is stable, controllable, and live-usable.

Statistical Truth from 10 Independent Trials

This report is based on 10 independent trials, each using the same road data but with Monte Carlo algorithm randomness. Bankruptcy rate 0/10 (0%), average net P&L +29,658 ¥ (+296.58%). This report showcases the highest-yield trial (T4).

Trial	Net P&L (¥)	Return	Max Drawdown	Bankrupt
T1	+54,514	+545.14%	58.7%	✓
T2	+19,495	+194.95%	83.7%	✓
T3	+925	+9.25%	94.2%	✓
T4 (this report)	+56,960	+569.60%	64.8%	✓
T5	+13,820	+138.20%	85.9%	✓
T6	+50,250	+502.50%	64.1%	✓
T7	+9,788	+97.88%	89.6%	✓
T8	+14,425	+144.25%	93.4%	✓
T9	+46,950	+469.50%	73.5%	✓
T10	+29,455	+294.55%	82.2%	✓
Average	+29,658	+296.58%	—	0/10 bankrupt

Key Findings from 10 Trials

Significant Engineering Progress

10/10 all profitable, bankruptcy rate 0/10 — vs v2.8.11's 5/6 profitable, T2-class bankruptcies completely eliminated
Average net P&L +29,658 ¥ (+296.58%), far above the v2.8.11 era
Worst trial (T3) still profitable at +9.25% — no v2.8.11-style -99.10% extreme cases
Median max drawdown ~80%, slightly better than v2.8.11's ~85%
Reverse Martingale stake (max_mult=4) + 5-algorithm ensemble significantly reduced tail bankruptcy risk

5-Algorithm Performance

Algorithm	Predictions	Hits	Actual Win Rate	Verdict
Hash Equation	283,259	144,652	51.07%	Break-even
Monte Carlo Sampling	283,259	142,690	50.37%	Break-even
Bayesian Inference	283,259	142,491	50.30%	Break-even
Chi-Square Deviation	263,259	132,323	50.26%	Break-even
Pattern Mining	265,421	131,405	49.51%	Loss

Stake Distribution (Reverse Martingale · max_mult=4)

Stake Amount	Usage Count	Share
100 ¥ (base unit)	34,174	49.79%
400 ¥ (4× cap on consecutive wins)	34,461	50.21%

* The 100/400 split shows Reverse Martingale stake works as intended: doubles up to 4× cap on win streaks (50%+ of hands), resets to base on losses (50% of hands).

Win / Loss Streak Statistics

Max Win Streak

10 hands

Longest during backtest

Max Loss Streak

16 hands

Longest during backtest

* 16-loss streak is what triggers the circuit-breaker (threshold 3). Each 3+ loss streak auto-pauses 5 hands to avoid larger tail risk.

Per-Shoe P&L Distribution

Category	Shoes	Share
Profitable shoes (profit > 0)	1,740	34.80%
Losing shoes (profit < 0)	3,215	64.30%
Break-even shoes (profit = 0)	45	0.90%
Total	5,000	100%

* 1,740 profitable shoes "winning big" covers 3,215 losing shoes "losing small" — this is the source of the +569% net P&L, the essence of Reverse Martingale stake.

Circuit Breaker Details

3-Loss-Streak → 5-Hand Pause

Circuit-breaker rule: when 3+ consecutive losses occur, the system auto-pauses betting for 5 hands (cool-down), with stake force-zeroed to prevent further losses. After cool-down, stake resets to base unit.

Metric	Value
Total circuit-breaker triggers	4,947
Avg per 100 shoes	98.94
Avg loss streak at trigger	3.00 hands
Earliest trigger	Shoe 2
Latest trigger	Shoe 5,000

Key Risks and Boundaries

What You Must Know

Backtest ≠ live. This report uses historical road simulation; live has slippage, emotion, execution gaps
While 10/10 non-bankrupt, v2.8.12 is still not a "guaranteed win" — 5000 shoes is a window, long-term still converges to negative expected value
Max drawdown 64.84% means: around shoe 3,531 you'll experience "near-zero" account state (though end profitable)
T3 trial only +9.25% — worst case you may "spend 5000 shoes only earning a little", must have psychological preparation
Baccarat is a negative-expectation game (banker −1.06%, player −1.24%), long-term −1.06%. Any backtest profit is a probability window not a steady state
Do not use living expenses or borrowed money. Only use fully-affordable-to-lose spare cash
18+ / local law permits / mental health — all three bottom lines required

Want to verify this backtest yourself?

PhD version offers 10-minute free trial. Telegram us to get the full evo_data.csv dataset + v2.8.12 algorithm open-source notes

📱 Telegram @baccapp ← Back to Home

Disclaimer: This backtest report is based on 5,000 shoes of real historical road data (313,259 hands), simulated using the Monte Carlo algorithm. Results are for methodological research and engineering validation only, not any investment advice, betting advice or profit guarantee. All live trading has slippage, psychology, execution gaps that may differ significantly from backtest. Baccarat is a negative-expectation game (banker -1.06%, player -1.24%), long-term returns converge to negative expected value. Do not treat gambling as a source of income. 18+ users, please gamble responsibly, comply with local laws. If you have a gambling addiction problem, please seek professional psychological help.

Report generated: 2026-06-10 14:41 (UTC+8) · Data source: evo_data.csv · Algorithm: v2.8.12 (max_mult=4) · Software: BaccAI