Blog: The Variance Filter Doesn't Care What Window You Use
Post-mortem on the variance-lookback-short-symmetric rejection (registered + tested + rejected 2026-04-17). Key points recorded for future-me:
Post-mortem on the variance-lookback-short-symmetric rejection (registered + tested + rejected 2026-04-17). Key points recorded for future-me:
- CLV is flat (+10.0-10.1%) across lookback in 5-10
- Lower lookback tightens the filter because the 3.0-goal threshold
is coupled to window length — shrinking the window makes +3 goals harder to accumulate, not easier
- The 22 early-season bets this experiment was motivated by are NOT
unlocked by lowering lookback alone; the real fix is a two-parameter window+threshold-scaling hypothesis that's a distinct test
- Second rejection in the variance-tuning space after
attack-defense-regression-asymmetry (rejected 2026-04-01). Together they show the variance filter's edge is insensitive to how the window is parameterised within normal ranges.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>