feat: B.League walk-forward — edge is NOT systematic (one-season artifact) - #13
Open
Manimall wants to merge 1 commit into
Open
feat: B.League walk-forward — edge is NOT systematic (one-season artifact)#13Manimall wants to merge 1 commit into
Manimall wants to merge 1 commit into
Conversation
Verifies whether B.League's single-season +19% ROI (CI95>0 on 142 in-season bets) is a real bookmaker inefficiency or a one-season fluke. - collector_config.py: add BLeague_2425 archive slug (Flashscore keeps historical O/U odds — confirmed a 2023 match still serves handicaps). - scripts/walk_forward_bleague.py: strict chronological expanding-window validation. Trains V6 (BASE, threshold 0.54, no tuning) on all prior seasons, bets the held-out season; per-season + pooled bootstrap-CI95 aggregate. - tests/test_walk_forward_season.py: guards the Aug season-rollover bucketing. Result (train 24/25 -> test 25/26, 688 bets): AUC 0.503, ROI -5.00%, CI95 [-11.90%; +2.18%]. The edge did NOT survive: cross-season AUC collapses 0.62 -> 0.50, ROI +19% -> -5%. The single-season "edge" was a within-season artifact (same teams/scoring environment), not a systematic inefficiency. Walk-forward caught a false positive on a 5x larger, cleaner sample. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Задача (Senior Quant)
Доказать, системная ли неэффективность букмекера эдж B.League (+19% ROI, CI95>0 на 142 внутрисезонных ставках) — или флуктуация одного сезона. Инструмент: строгий мульти-сезонный walk-forward.
Что сделано
collector_config.py), собрано 736 матчей сезона 24/25 + odds-бэкфилл 100% (736/736). Подтверждено: Flashscore хранит исторические O/U (матч 2023 г. отдаёт handicap'ы).scripts/walk_forward_bleague.py: строго хронологический expanding-window. V6, только BASE, порог 0.54, никаких подгонок. Обучение на всех прошлых сезонах → ставки на отложенный сезон; per-season + пул-агрегат bootstrap CI95.Результат: train 24/25 → test 25/26
❌ Вердикт: эдж НЕ системный
Грааль не пережил кросс-сезонный тест:
Почему: прежний +19% использовал сплит ВНУТРИ одного сезона (те же команды, та же скоринговая среда) → модель ловила внутрисезонную структуру, которая не переносится на модель с прошлого сезона. Ранжирующая способность не генерализуется между сезонами — значит системного эджа нет.
Ценность
Walk-forward отловил ложное срабатывание на большой чистой выборке — уберёг от ставок на фантомный эдж. Тот же класс ошибки, что ловила слепая валидация, но теперь по объёму данных. Инструмент переиспользуемый (
--расширяется на N сезонов).Гейт
ruff/mypy чисто · pytest зелёные · файлы ≤250 строк · строгая типизация · dataclass · без print/hardcode.
🤖 Generated with Claude Code