Shadow cohort 적중률
v1.8 (production) vs v2.1-B-shadow 가중치 일별 Brier + 적중률 delta. 동일 경기 동일 input 으로 quant 재계산 (debate LLM 호출 X, 비용 0). n=150 임계 달성 완료 — 가중치 v1.8 유지 확정.
내부 evidence cohort 입니다. 실제 예측은 예측 기록 에서 확인하세요.
누적 표본 (n)
192
v1.8 평균 Brier
0.2462
v2.1-B-shadow 평균 Brier
0.2455
delta -0.0007
Shadow 실험 가중치 구성
sp_fip
16%
sp_xfip
5%
lineup_woba
17%
bullpen_fip
11%
recent_form
12%
war
9%
head_to_head
2%
park_factor
4%
elo
9%
sfr
0%
park_weather
3%
umpire_sz
2%
shadow factor: park_weather 3% · umpire_sz 2% — production weight 0 양립.
Calibration plot (10-bucket)
예측 확률 vs 실제 적중률. 대각선 위 = 잘 맞은 bucket, 대각선 아래 = 과신 (over-confident). v1.8 (production) vs v2.1-B-shadow (shadow) 두 모델 calibration 차이 시각화.
일별 cohort (최대 14일)
| 날짜 | n | v1.8 Brier | v2.1-B-shadow Brier | delta | v1.8 적중 | v2.1-B-shadow 적중 |
|---|---|---|---|---|---|---|
| 2026-07-18 | 5 | 0.2613 | 0.2602 | -0.0011 | 2/5 (40.0%) | 2/5 (40.0%) |
| 2026-07-17 | 4 | 0.3063 | 0.3022 | -0.0041 | 0/4 (0.0%) | 0/4 (0.0%) |
| 2026-07-16 | 5 | 0.2618 | 0.2715 | +0.0097 | 2/5 (40.0%) | 2/5 (40.0%) |
| 2026-07-09 | 4 | 0.2349 | 0.2295 | -0.0053 | 2/4 (50.0%) | 3/4 (75.0%) |
| 2026-07-08 | 4 | 0.2537 | 0.2265 | -0.0272 | 3/4 (75.0%) | 3/4 (75.0%) |
| 2026-07-07 | 5 | 0.2516 | 0.2589 | +0.0073 | 2/5 (40.0%) | 4/5 (80.0%) |
| 2026-07-05 | 3 | 0.2514 | 0.2334 | -0.0180 | 2/3 (66.7%) | 2/3 (66.7%) |
| 2026-07-04 | 5 | 0.2405 | 0.2447 | +0.0042 | 3/5 (60.0%) | 2/5 (40.0%) |
| 2026-07-03 | 5 | 0.2549 | 0.2603 | +0.0053 | 2/5 (40.0%) | 1/5 (20.0%) |
| 2026-07-02 | 5 | 0.2269 | 0.2254 | -0.0015 | 4/5 (80.0%) | 4/5 (80.0%) |
| 2026-07-01 | 5 | 0.2918 | 0.2742 | -0.0176 | 0/5 (0.0%) | 1/5 (20.0%) |
| 2026-06-30 | 4 | 0.2215 | 0.2406 | +0.0190 | 3/4 (75.0%) | 2/4 (50.0%) |
| 2026-06-28 | 5 | 0.2143 | 0.2257 | +0.0114 | 4/5 (80.0%) | 4/5 (80.0%) |
| 2026-06-27 | 5 | 0.2805 | 0.2702 | -0.0102 | 1/5 (20.0%) | 2/5 (40.0%) |
본 페이지는 shadow cohort 누적 진단용입니다. v1.8 production 가중치 변경 X.