Internal deployment, dated 2026-09-01/02. Sources: F4221 fill-mode taxonomy, F4292 contamination audit.
Before offering Fidelity Lab's backtest methodology to customers, we ran it against our own ~42-million-leg internal trading ledger. The audit found the ledger's "real_fill=1" flag was a typed literal written by the emitter on every tape-priced paper leg — every one of 12.3 million "clean real-fill" legs was actually priced off historical tape with a simulated fill, not a genuine venue-reported execution. A separate forward-paper check found roughly 95.8% of positive top-line results before the fix traced to lookahead contamination, caught by an 11-detector battery (fill-mode mislabeling, epoch-zero entry timestamps, settlement-clock field mismatches, and more) built specifically to keep a backtest PIT-faithful. We fixed the labeling (a `fill_mode` column: venue_fill vs tape_priced vs synthetic) and re-ran every verdict split by tier before trusting any of it — the same walk-forward, cost-netted, per-market-ruler discipline every Fidelity Lab backtest runs for a customer's own strategy.