Our strongest signals were our worst trades — the audit proved it

We let our practice auto-trader place its first small trades, then audited the results honestly. The strength score was backwards: the highest band, trades scored 80 and up, had zero wins in 41 trades. We pre-registered a redesign before touching the code, and the redesign failed its own test too. Version 2 backtested worse per trade, so we left it disarmed and dry-running. It stings, but that is the point of practice money and a written audit — you find out your best-looking signal is your worst one before real money ever learns it for you.