Rolling-origin backtest เทียบ random baseline
แต่ละงวดใช้เฉพาะข้อมูลก่อนงวดเป้าหมาย สร้าง ranking แล้วตรวจผลจริงงวดถัดไป เพื่อลด look-ahead leakage
User-defined backtest
| Strategy | Top K | Evaluated draws | Hits | Hit rate | Random baseline | Lift | 95% CI | สรุป |
|---|---|---|---|---|---|---|---|---|
| Strategyhot-20 | Top K5 | Evaluated draws724 | Hits34 | Hit rate4.70% | Random baseline5.00% | Lift-0.30% | 95% CI3.38% ถึง 6.49% | สรุปNo clear evidence of edge over random baseline. |
| Strategyhot-50 | Top K5 | Evaluated draws724 | Hits35 | Hit rate4.83% | Random baseline5.00% | Lift-0.17% | 95% CI3.50% ถึง 6.65% | สรุปNo clear evidence of edge over random baseline. |
| Strategyhot-100 | Top K5 | Evaluated draws724 | Hits29 | Hit rate4.01% | Random baseline5.00% | Lift-0.99% | 95% CI2.80% ถึง 5.69% | สรุปNo clear evidence of edge over random baseline. |
| Strategycold | Top K5 | Evaluated draws724 | Hits29 | Hit rate4.01% | Random baseline5.00% | Lift-0.99% | 95% CI2.80% ถึง 5.69% | สรุปNo clear evidence of edge over random baseline. |
| Strategyall-time-frequency | Top K5 | Evaluated draws724 | Hits26 | Hit rate3.59% | Random baseline5.00% | Lift-1.41% | 95% CI2.46% ถึง 5.21% | สรุปObserved historical result was below random baseline. |
| Strategybalanced | Top K5 | Evaluated draws724 | Hits30 | Hit rate4.14% | Random baseline5.00% | Lift-0.86% | 95% CI2.92% ถึง 5.85% | สรุปNo clear evidence of edge over random baseline. |
| Strategydigit-position | Top K5 | Evaluated draws724 | Hits37 | Hit rate5.11% | Random baseline5.00% | Lift0.11% | 95% CI3.73% ถึง 6.96% | สรุปNo clear evidence of edge over random baseline. |
| Strategypair-based | Top K5 | Evaluated draws724 | Hits26 | Hit rate3.59% | Random baseline5.00% | Lift-1.41% | 95% CI2.46% ถึง 5.21% | สรุปObserved historical result was below random baseline. |
| Strategyuniform-random | Top K5 | Evaluated draws724 | Hits31 | Hit rate4.28% | Random baseline5.00% | Lift-0.72% | 95% CI3.03% ถึง 6.01% | สรุปNo clear evidence of edge over random baseline. |
Leakage guard
Each evaluation uses only draws with draw_date before the evaluation draw. All qualified rolling-origin draws are evaluated.
ที่มาและความน่าเชื่อถือของข้อมูล
- Sample size
- 744 งวด
- Total exposure
- 744
- Dataset version
- d20261016.744
- Methodology version
- statistical-methodology-v1.0.0
- ช่วงข้อมูล
- 1995-02-01 ถึง 2026-10-16
- Computed at
- 2026-10-16T00:00:00.000Z
ข้อจำกัด: Past frequency describes historical results only and does not predict future draws. Eligibility and exposure are evaluated per prize type; missing unrelated tiers do not exclude verified headline prizes. Backtest results describe historical strategy behavior only and do not establish future performance.