Skip to content

Validation dossier, Alpha Readiness Level and live monitoring

Three tools for the steps after a backtest is verified: putting it before a committee, saying where it stands on the way to capital, and watching it once it trades.

monte-neo dossier: one folder for a committee, a validator or an auditor

monte-neo verify then monte-neo dossier --framework sr11-7: the folder a committee can file

monte-neo dossier --demo --framework sr11-7                 # no files
monte-neo verify --ohlcv data.csv --strategy s.py --out cert.json
monte-neo dossier --certificate cert.json --framework ss1-23 --name "Momentum 20" --owner "Desk A" \
  --purpose "Daily futures overlay" [--alpha alpha.json] [--live live.json] --out dossier/

The folder holds:

  • dossier.html, one page with no scripts and no network, printable to PDF: the verdict, purpose and scope, the results, every check, the Alpha Readiness Level, the framework map, the assumptions and limitations, how to reproduce, and blank sign-off fields;
  • dossier.json (validation-dossier/1), the certificate itself, environment.json (Python, platform, package versions), REPRODUCE.md (the exact command and the expected certificate id) and limitations.md.

Frameworks (--framework):

Name Framework
generic Data, look-ahead, economics, statistics, claims, policy, reproducibility, monitoring
sr11-7 Federal Reserve SR 11-7 / OCC 2011-12, model risk management
ss1-23 PRA SS1/23, model risk management principles for banks
rts6 MiFID II RTS 6 (Delegated Regulation 2017/589), testing of algorithms
crypto A checklist for crypto funds and desks: 24/7 data, funding, venue latency, delisted tokens, crash stress
academic A paper's appendix: pre-registration, trials disclosed, out-of-sample, the t ≥ 3 hurdle, reproducibility

--lite also writes dossier-lite.html, one plain-language page for a reader who is not a quant (a family office, a board), with the notice in a box at the top.

The map shows, per clause, whether the dossier holds evidence for it (covered, partial, not covered) and what that evidence says. Clauses that need a person, such as conceptual soundness or governance, are never shown as fully covered. A dossier is evidence for a validation, not a statement that anyone complies with a framework.

MCP tool: build_dossier.

Alpha Readiness Level (ARL 1–9)

Every alpha report and every dossier carries an ARL: where the evidence stands on the way from an idea to capital.

ARL Level Evidence
1 Idea on clean data data_integrity, or an aligned scores panel
2 The backtest is not broken look-ahead probes, or universe look-ahead; any failed look-ahead check stops here
3 It earns after costs net_profitability, or market_impact at your --aum
4 The signal predicts ic and signal_placebo, or timing_significance
5 It is new alpha residual_alpha (t ≥ 3 after factors) and no hidden_short_volatility
6 It survives the search deflated_sharpe, or the t ≥ 3 hurdle with enough history
7 Tradable and robust impact or capacity; synthetic markets, subsets and start dates (or period consistency and delays); rebalance day (or cost margin)
8 It adds to your book marginal_sharpe with --book
9 Confirmed live live_vs_backtest and alpha_death_monitor

The level is the highest one met together with every level below it. Missing evidence is not assessed, never passed, and the report names the run that would produce it, for example ARL 6/9 — blocked at 7: tradable and robust (rebalance_timing_luck).

The ARL rates the evidence, not the investment.

monte-neo monitor: after launch

monte-neo monitor --certificate cert.json --live live.csv [--fills fills.csv] [--haircut 0.5] [--out live.json]

live.csv has live or paper returns (date, return). fills.csv has timestamp, symbol, side, quantity, price and, for slippage, the decision price reference_price.

Check What it says
live_vs_backtest Correlation, tracking error and the mean difference against the backtest on the same dates
implementation_shortfall Realised slippage against the modelled one, and the value to use instead
alpha_death_monitor Sequential alarms (CUSUM, SPRT) against the backtest edge shrunk by --haircut, with the date of the alarm
live_track_needed How much live history is still needed to tell the edge from zero

Exit code 0 means no warning and 2 means a warning, so it can run from cron or CI. MCP tool: monitor_live.

Not investment advice: a research example, not a call to trade or a promise of profit. Simulated results do not predict future returns.