LEDGER
// before you go live

Your backtest is lying to you. Find out how, before your broker does.

A command-line toolkit that puts your trading strategy through the same walk-forward validation, realistic cost modeling, and look-ahead-bias audit that professional quant desks use — before you risk real capital finding out the hard way.

REJECTED
backtest_v1.pysingle split
Total return+38.4%
Sharpe1.9
Win rate61%
flat spread · no purge window · single train/test split
ledger walk-forward13 windows
Total return-2.2% HONEST
Sharpe-0.59
Windows profitable2 / 13
per-instrument cost model · 20-bar purge · TIER-based exits simulated

A real (anonymized) result. The strategy above looked ready to trade live. It wasn't — and the only way to know was to check.

What it actually checks

Not a magic signal detector. A skeptic in software form — it looks for the specific, well-documented ways backtests lie.

▣
Single-split overconfidenceForces rolling, non-overlapping out-of-sample windows. One good backtest becomes thirteen honest ones.
▣
Underpriced execution costsPer-instrument spread and slippage, not one flat number applied to everything from majors to illiquid pairs.
▣
Look-ahead leakageAudits every feature for information that couldn't have existed at decision time — the multi-timeframe resample bug that inflates AUC to 0.80 and feels like genius.
▣
No purge between train and testFlags rolling windows and label horizons that let future information bleed backward across the split.
▣
Exit logic that never actually runsChecks whether your simulated stop-loss/take-profit/trailing logic would ever trigger — or just looks like it does.
▣
Multiple-comparison blindnessScanning 40 features across 6 instruments? FDR-corrects your significance claims so "found something" isn't just arithmetic.

How it works

1
Point it at your strategy
Your existing signal/entry logic and historical data. Python, no rewrite required.
2
It runs the audit
Walk-forward simulation, cost stress tests, and a static look-ahead-bias scan of your feature code.
3
You get a verdict, not a vibe
Per-window results, not one aggregate number. If there's no edge, the report says so — a well-validated "no" is a successful outcome, not a bug.

Get it when it ships

This doesn't exist yet. I'm building it if enough people confirm they'd actually use it — the same discipline the tool itself is built on: check before you commit.

$300 one-time, no subscription

Full toolkit, lifetime updates for the version you buy. Not a strategy. Not a signal. A second opinion your future self will thank you for.

No spam, no drip campaign — one email when it's ready, or when I decide it isn't happening and tell you why.