Concept · Flat mode (Fixed orders)
Position size = constant dollar amount per trade, regardless of account balance. Isolates signal quality from capital-scaling dynamics. Structurally superior for measuring whether a strategy's edge is real.
When flat mode is on, every trade uses the same dollar size — $1,000 per trade whether your account is up 50% or down 30%. This doesn't reflect how real trading accounts typically work, but that's the point: by removing the compounding effect, the P&L curve becomes a direct readout of signal quality.
Win a trade → your account grows by $X. Lose a trade → your account shrinks by $X. The distribution of trade outcomes is path-independent — it doesn't matter whether wins happened early or late.
positionSizeUsd = orderSizeUsd (constant, set at config time)
EquityLedger is bypassed entirely in flat mode. No bankruptcy tracking. No compounding. EquityLedger.recoverFromTrades() short-circuits immediately.
UI label: Fixed orders. Config key: equityMode: false.
"Is the underlying signal — the pattern of entry/exit conditions — actually predictive of price direction, independent of how I size my trades?"
This is the cleanest measurement surface for signal quality. Because position size never changes, every trade's outcome contributes equally to the Sharpe calculation regardless of when it occurred.
1. No left-tail truncation In equity mode, bankruptcy halts the sim and clips the worst losing streaks from the return distribution, inflating Sharpe. Flat mode has no bankruptcy — every trade, including brutal losing streaks, enters the distribution. The Sharpe is computed on the full return history.
2. Path-independence In equity mode, lucky early wins grow future position sizes and amplify later wins — even if the signal quality is identical to another strategy with unlucky early runs. Flat mode: constant size → the return sequence doesn't compound → signal quality is isolated. Two strategies with the same signal skill rank the same in flat mode regardless of win/loss ordering. See path dependence.
3. Stronger leaderboard comparability The comparability invariant (one backtest window for all strategies) is cleaner in flat mode. No strategy benefits from compounding amplification during their particular win/loss sequence within the shared window.
Flat mode IS a poor model of real account dynamics. Real accounts compound. Real traders face bankruptcy risk. Flat mode does not simulate these.
But "unrealistic" doesn't mean "less useful" — it means flat mode answers a different question than equity mode. For signal quality measurement, the unreality is the point: you're controlling for capital dynamics to isolate the signal.
Use equity mode when you want to know what would happen to your account.
Use flat mode when you want to know if the signal is any good.
If equity mode says "Real edge" but flat mode says "Inconclusive" for the same strategy: the equity verdict is suspect. The most likely culprits are leverage mirage (bankruptcy truncated the worst losses, inflating Sharpe) or path dependence (early wins happened to compound into larger positions).
Re-run the strategy in flat mode before trusting an equity-mode "Real edge" at high leverage.
apps/backend/src/simulation/config-manager.ts:241 (resolvePositionSize)apps/backend/src/evaluation/equity-ledger.ts:156 (recoverFromTrades no-op in flat mode)apps/backend/src/evaluation/evaluator.ts:1721–1724, 1879–1883 (equity-mode gating)wiki/qa-sessions/2026-07-01-session.md#q1Related concepts
See it in a real result →Put it to the test
Spawn your variant, run it on the same engine, and read the edge-significance verdict — before you risk real money.