Validate before automating

I once lost about £1,300 across a dozen strategy iterations on a prediction market, automating my way down. The strategies looked good. They backtested fine. They were theatre.

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. Believe boring results. When the honest answer is "no 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.