Case 1 · Look-ahead
Case 1 — Look-ahead: the exit scanned the future on the entry bar
Symptom. A single-asset perpetual strategy backtested beautifully: high win rate, steady equity curve. It produced almost nothing live. The edge was a future leak.
Root cause. The signal generator decided whether a trade would hit its stop-loss or take-profit by scanning a forward window at the moment of entry:
for ei in range(n):
if not condition(df, ei):
continue
hit = None
for j in range(ei + 1, min(ei + 1 + TO, n)): # <-- scans the FUTURE
if low(df, j) <= sl_price(df, ei):
hit = "SL"; break
if high(df, j) >= tp_price(df, ei):
hit = "TP"; break
if hit == "TP": # only trade the winners
entries.append(ei)
Every emitted signal already knew its own outcome. auditkit.checks.lookahead
flags this as FUTURE_WINDOW_LOOP (severity HIGH, because the loop is nested
inside the outer bar loop).
What the fix changed. Evaluate the exit only on the current bar, as the
backtest advances — no decision on bar i may see bar i+1.
| Variant | Trades | Win rate | Net PnL |
|---|---|---|---|
| Bar-level scan of the future (buggy) | 91 | 90.1% | +$214.37 |
| Current-bar scan only (fixed) | 26 | 65.4% | -$0.12 |
| Tick-level scan only (fixed) | 27 | 66.7% | +$0.37 |
Bottom line. The leak was worth roughly $214 of imaginary PnL — the entire reported profit. Fixing it turned a "90% win-rate winner" into a break-even strategy. No parameter change; only the information available at decision time.
How to reproduce. tests/fixtures/lookahead_buggy.py triggers the detector;
tests/fixtures/lookahead_fixed.py is clean.