Following this capper-tracking system as a betting signal would have lost money across every testable strategy over the six-month window studied. The worst outcome is the fade-the-most-negative-losers approach: rather than printing a contrarian profit, fading the broad pool of proven-negative cappers produced a −5.5% ROI (−31u on 572 bets, CI fully below breakeven at the −20u threshold variants), which is meaningfully worse than simply betting every pick blind at −1.2%. The most enticing surface signal — three named cappers (ThisGirlBetz, Underdog Sniper, MrBigBets) faded at +17.0% ROI with a CI that barely clears zero — is disqualified by sample size (116 bets, minimum required is 200) and carries a hard selection-bias warning: those three names were chosen by looking at the full history before the backtest ran, making the "walk-forward" qualification a necessary but not sufficient safeguard against data-mining. There is one mechanical survivor (NBA-only bets, ROI +14.1%, p≈0.0005), but it was independently flagged as a likely data artifact: a 59.6% hit rate over a single playoff cycle, driven almost entirely by April–May, with June collapsing to 51.5% — the pattern of a hot-streak cohort entering the dataset during playoff month, not a repeatable edge. After the red-team audit confirmed one genuine look-ahead leak (the consensus strategy was peeking at same-week picks; fixed, reducing consensus bet count to zero), nothing survives with a publishable edge. The honest product of this backtest is not alpha — it is a clean, audited data infrastructure and a CLV-tracking pipeline now running as the ongoing truth metric.