Validate before automating

Earlier project notes described losing money across repeated prediction-market strategy iterations despite encouraging backtests. The precise historical loss total is not independently verified here. The failure, rather than the flattering simulation, is the important part of the account.

The lesson, paid for in full: validate before automating.

Concretely, that means:

1. Measure on settled outcomes, not on a backtest you tuned until it smiled at you.
2. Forward paper-trade against live data first — especially where no historical odds exist to backtest against.
3. Set a gate before any money or automation: e.g. "I will not auto-act until the manual version clears a real, pre-registered threshold."
4. Accept negative and inconclusive results. When the evidence does not support an actionable edge, the correct action is to stop — see the crash-game write-up.

It's the same discipline behind expected value and Kelly criterion: a machine will execute a bad idea faster and more reliably than you ever could by hand. Earn the automation. Don't assume it.