A strategy can be statistically robust and still fail when capital scales.
Capacity is the amount of capital your system can deploy before execution quality degrades unacceptably.

Why capacity matters

As size increases, you often get:

worse fills,
higher slippage,
longer time to complete orders,
higher exposure during execution.

At some point, net edge decays faster than gross return grows.

Capacity is pair- and regime-dependent

Capacity is not one global number. It depends on:

instrument liquidity profile,
trading session and volatility regime,
order type and urgency,
participation rate constraints.

A strategy might scale on BTC/USDT but break on alt pairs.

Practical way to estimate capacity

Use a staged curve instead of a single estimate:

Start from current notional per trade.
Simulate 2x, 4x, 8x sizing with stronger slippage assumptions.
Recompute net edge and drawdown per stage.
Stop where risk-adjusted performance crosses your minimum bar.

That crossing point is your operational capacity boundary.

Signs you are above safe capacity

PnL dispersion rises while signal quality is unchanged.
Win rate remains stable but average trade deteriorates.
Slippage variance dominates model edge.
Drawdown accelerates after size increases.

These are execution symptoms, not necessarily model-quality symptoms.

Capacity-aware deployment policy

Use pair-level capital caps.
Reduce size under spread/volatility stress.
Route larger orders in slices, not bursts.
Define kill-switch triggers for execution drift.

Without capacity controls, scaling can invalidate otherwise good research.

Maker versus taker changes the capacity curve

Taker-heavy strategies hit the book harder and decay faster with size.

If your backtest assumes mid fills, capacity estimates will be fantasy.

Toxic flow: your fills worsen when you are right

Some edges show up as adverse selection: you buy before further adverse short-term drift.

Stress capacity with slippage models that widen when volatility spikes, not only when size doubles.

Measure participation, not only notional

A useful operational metric is your share of recent volume during your typical hold window.

If that share is creeping up over months, you are drifting toward a liquidity collision even if notional looks stable.