Research reference
Worked cases
Small enough to reconcile by hand. Explicitly synthetic inputs, with the full calculation exposed.
From a frozen score to a reconciled synthetic ledger
Follow six invented securities through dated availability checks, a top-three allocation, return drift and three compounded teaching periods. The arithmetic is reproducible; it is not a market backtest.
Cost sensitivity without changing the selected securities
Keep the invented scores and allocation rule fixed, vary only the per-leg cost assumption, and reconcile pre-fee turnover with actual fee-funded traded notional.
Reject a future signal before it reaches the ledger
A seemingly harmless one-second timestamp change invalidates the synthetic formation. The engine fails the whole dossier instead of dropping a row or treating an unavailable observation as known.
A split, a distribution and the securities that stay in the sample
Separate raw price changes from split-adjusted price and total returns, while keeping the fixed six-security universe intact. A missing loser is an input failure, not a harmless cleanup.