Most backtest tools show you a curve and ask for trust. AxiomStrat's release process uses an independent verification harness whose only job is to catch accounting or execution-model errors.
The verification harness is a completely separate program. It reads the raw price bars and the trade list a backtest produced — and then rebuilds every trade by itself, from scratch, using nothing from the engine's code. If the engine cheated, fudged a fill, mis-priced a partial, or forgot a commission, the two versions won't match — and the harness fails loudly.
Every rule is checked both ways. Entry price = bar close ± half spread ± slippage. Stop-loss checked before take-profit. Partial closes booked at exactly round(vol × pct, 2). Commission charged on entry and on every closing fill, including partials and the final remainder. Nothing is assumed; everything is re-derived and compared to the engine's own records.
Same inputs, same result — twice in a row, bit for bit.
Symbol, timeframe, dates and count match what was loaded.
Every trade independently rebuilt from raw bars.
Full equity curve matches the replay to ±$0.03.
Final balance = starting balance + net P/L, with costs already included exactly once.
Commission, slippage and spread summed independently on every entry volume.
Reported per-day and per-month P/L add up to the totals.
Each trade's net P/L lands on the day it becomes realized — no date shuffling.
Win rate, profit factor, drawdown, Sharpe, SQN… all recomputed independently.
Paths, bands and probabilities stay consistent with the equity input.
Every train exit happens before every test entry. No leakage.
Multi-partial exits replayed and reconciled — volume, price and commission.
Verification isn't one happy-path backtest. The harness runs five scenarios built to stress every mechanism the engine has:
| Scenario | What it stresses | Result |
|---|---|---|
| RSI partials + full costs | Partial take-profits, commission, slippage, spread, risk % sizing | 322 trades replayed |
| RSI compounding + guardrails | Compound sizing, daily DD limit, trade caps, session windows | 86 trades replayed |
| MACD, no costs | Clean fills and simple exits | 191 trades replayed |
| Grid, concurrent positions | 8 positions at once, forced closes at end-of-data | 81 trades replayed |
| Mean Reversion, M5 | Multi-trade momentum entries on fast data | 198 trades replayed |
And the harness found real bugs. This is the point: it caught an end-of-data bug that left phantom open positions, a partial-take-profit bug that erased original stop/target levels, and a template bug that crashed on certain strategies. Each one was fixed, and the harness now passes 62/62 on every release.
The independent verification harness is run against real MT5 history as part of release QA. Version 0.5.3 passes all 62 checks across five strategy scenarios.
Download the verified build