Monte-Neo¶

The independent verifier for trading strategies written by AI agents and humans
Catch look-ahead bias, hidden trading costs and overfitting before a backtest reaches your money

Install

pip install monte-neo
monte-neo verify --ohlcv prices.csv --strategy my_strategy.py --n-trials 12

The problem¶

A coding agent can turn a trading idea into a backtest in minutes, and then report a Sharpe of 3. Most of the time that number is wrong:

Look-ahead bias: the code reads future bars (shift(-1), centred windows, bfill, whole-series statistics, np.gradient, FFT filters).
Missing costs: the edge is smaller than fees and slippage, or needs a perfect fill.
Selection bias: the best of 300 variants is reported as if it were the only one.

What Monte-Neo does¶

It checks the backtest, not the idea. Give it the price data and the strategy code or its positions. It returns PASS, PASS_WITH_WARNINGS, NEEDS_MORE_EVIDENCE or REJECT, the checks behind the verdict, concrete next steps for the agent, and a reproducible strategy-verdict/1 certificate that can be signed.

Family	Checks
Look-ahead	Truncation and future-perturbation probes, outside-data watch, static AST lint (23 rules), implausible hit rate
Economics	Net return after costs, break-even cost in bps, one- and two-bar execution delay
Statistics	Probabilistic and Deflated Sharpe priced by `n_trials`, Monte Carlo timing test, sample size, holdout; walk-forward and parameter plateau for grids
Integrity	Broken OHLCV (NaN, bad prices, bars out of time order), non-deterministic signals

New in v0.53: from a rejected backtest to a committee

Repair without overfitting: defects fixed, choices made by the number of defects on 70 % of the history, a lockbox opened once, and an honest REPAIRED_NO_EDGE when the profit was the defect.
Alpha validation with an Alpha Readiness Level: signal power, factor attribution, capacity, robustness, value to your book.
Validation dossier and live monitor: evidence mapped to SR 11-7, PRA SS1/23 and MiFID II RTS 6; alarms when the edge dies after launch.
ML models and event data, track records from returns, policy, tracker trials and LLM cutoffs, a reference alpha.

New: catch look-ahead that hides in milliseconds¶

monte-neo verify --quotes

A quote has two times: when the exchange stamped it and when it reached you. A backtest that bins by the stamp lets a fast strategy trade on prices it could not have seen yet, and no code lint can find that. monte-neo verify --quotes runs the strategy on the time the data arrived, names the delay in ms where its profit vanishes, redraws the observed latency 200 times and issues a certificate you can reproduce. Read the guide.

Where to use it¶

Inside your coding agent: Claude Code plugin, or the MCP server for Codex, Gemini CLI and Cursor. See Use from agents.
In CI: a GitHub Action that fails the pull request on REJECT and posts the verdict.
When reviewing someone else's strategy: re-check their certificate and verify its signature.
When screening submissions at a prop firm, marketplace or course.
When comparing agents: the Honesty Bench.

Learn more¶

Verifier API: checks, verdicts, certificates, re-checks and signatures
Frameworks and notebooks: vectorbt, Freqtrade, Lean, Zipline, Jupyter, pre-commit, badge, Docker
Trap Suite: 80 strategies that lie and 35 honest controls, and how to add yours
Honesty Bench: score how honestly agents report backtests
Verify a certificate: check a signed certificate in your browser
Six ways your agent's backtest lies: the traps, with numbers
FAQ · Changelog · Roadmap

Research engine (maintenance mode)¶

Monte-Neo started as a fast local research engine for Apple Silicon: fee-aware next-bar backtests, parameter sweeps with golden vectors, Monte Carlo helpers and a paper OMS, with Metal / MLX / Numba device selection. The verifier runs on this engine. The engine still works and receives bug fixes, but new work goes into the verifier.

from monte_neo.backtest import ExecutionModel, export_sma_sweep, synthetic_ohlcv

ohlc = synthetic_ohlcv(100_000, seed=42)
model = ExecutionModel(commission_bps=5.0, slippage_bps=5.0, warmup_bars=50)
out = export_sma_sweep(ohlc["open"], ohlc["high"], ohlc["low"], ohlc["close"], combos=16, model=model, device="auto")
print(out["device"], out["combos"])

See the quick start and the export API.

SMA sweep demo

Memory plan demo

Source: github.com/NeoZorK/Monte-Neo · MIT license · Not investment advice.