How long paper trade before live and paper trading before live trading are among the most common deployment searches. The honest answer is not a fixed number of days. Paper trading tests operations: connectivity, order types, latency, and whether your signals behave like research when infrastructure is real.

What paper trading does not prove

Paper trading does not guarantee profit because fills are not fully adversarial. It also does not fix overfitting in research (Overfitting guide).

What paper trading is good at

Catching API and config mistakes early
Verifying risk limits and position sizing logic
Building logging habits before capital is on the line

Practical duration guidance

Run paper long enough to see multiple regime slices for your timeframe (weeks, not hours for most swing systems). If you change parameters or code mid-run, reset the clock or label the change.

When to move to live with small size

Move when you have stable logs, a kill-switch plan (Strategy kill-switch), and validation evidence that matches your risk tolerance (Deploy framework).

Paper-trade metrics that should look boring

Good paper runs produce boring telemetry:

stable order latency distribution
stable cancel/replace counts near zero
stable trade frequency within an expected band versus OOS

If these drift while PnL still looks fine, you are building a time bomb.

When paper trading should stop (even if you want more comfort)

Stop extending paper indefinitely if:

you keep changing parameters to "fix" paper PnL
you are avoiding live because of fear, not because of missing evidence

At that point, go live small or redesign the strategy. Paper cannot resolve fear.