Same-bar execution
LOOKAHEAD_OR_SAME_BAR_ENTRY
Flags execution timestamps that are not causally later than the signal bar close.
Install in one command
python -m pip install backtest-integrity-guardCurrent release: 0.2.2 · Python 3.10+ · MIT · no runtime dependencies.
Four failure modes worth checking first
Missing bars
MISSING_BARS
Checks a declared cadence and reports undeclared gaps instead of silently accepting them.
Ambiguous exits
AMBIGUOUS_SAME_BAR_EXIT
Fails closed when stop and target both hit inside one OHLC bar without an explicit policy.
Timestamp conflicts
TIMESTAMP_CONFLICT
Distinguishes an exact duplicate bar from same-timestamp rows that disagree on OHLC(V).
Quick audit
# OHLCV integrity + exact input hash
btguard ohlcv bars.csv --interval-seconds 300 --hash-input
# Signal-to-execution causality
btguard ledger trades.csv --hash-input
# Freeze research inputs
btguard freeze bars.csv trades.csv -o manifest.json --root .
btguard verify manifest.json --root .Designed for explicit, reproducible assumptions
Deterministic reports
Stable JSON schema, tool version, deterministic serialization, and optional SHA256 binding to the audited file.
Vendor schemas
Map vendor-specific timestamp/OHLCV column names without silently guessing missing fields.
Explicit session gaps
Allow known breaks intentionally while keeping undeclared gaps fail-closed.
Framework agnostic
Works on CSV exports from custom research code or existing backtesting engines.
What it does not claim to solve
It does not model commissions, slippage, market impact, exchange calendars, order-book fills, survivorship bias, or strategy overfitting. The scope is narrower: make several common research-integrity assumptions explicit, testable, and reproducible.