Check validity and economics
Promotion criteria examine issues such as holdout behavior, overfitting, sample adequacy and trading costs. Findings stay attached to the run.
Risk & release review
A result needs review before release.
AI analysis, risk and portfolio agents review the strategy alongside enforced checks. Paper admission requires recorded approval; supported live execution also requires your explicit authorization.
See the strategy factory in motion.
Illustrative example · not a live result
Watch AI agents build and review a strategy.
AI agents develop the idea, write code and assess the test results. Review findings can send the strategy back for another iteration.
Watch the path, not just the endpoint.
An illustrative equity curve reveals setbacks along the way. Inspect drawdown and trading costs before interpreting a result.
Keep open positions in the picture.
Compare account equity with closed-trade balance. An open position can change equity before a trade closes.
- Describe
- Build & verify
- Backtest
- Stress-test
- Paper-trade
EquityClosed-trade balanceDrawdown
Net return—
Max drawdown—
Checkpoint—
Review exceptions explicitly
Some quality criteria allow a documented AI analyst waiver. A separate AI risk reviewer examines each proposed exception and can confirm or reject it.
Record the release decision
Admission and release checks control paper-trading entry. Review artifacts provide context for the decision and for later monitoring.
Follow the decision, not just the score.
A strong in-sample metric alone is not permission to deploy. Read the AI analyst’s findings, any waivers, the AI risk reviewer’s response and the recorded admission outcome together.
Assessment
Validity, economics, quality
Independent review
Risk findings and exceptions
Admission
Recorded release outcome
Understand the loss behind the return.
A strategy can finish positively and still spend a long time recovering. Read drawdown depth, recovery time and position sizing together.
Inspect whether losses cluster in a market regime, a small group of trades or concentrated exposure. A single headline risk metric cannot explain the full path.
Peak: 100Trough: 88Recovery: 100
Different findings need different decisions.
The review separates blocking validity failures, quality exceptions and advisory findings. Read the classification recorded for the run.
Scroll horizontally to see all columns.
| Finding | Meaning | Next step |
|---|---|---|
| Hard failure | An enforced validity check failed. | Resolve the issue; an analyst waiver cannot override it. |
| Quality exception | A reviewable criterion was missed. | Read the analyst justification and the independent risk response. |
| Advisory or unevaluated | A finding is informational, in rollout, or lacks a required metric. | Check what was and was not established; absence is not evidence of a pass. |
Approval belongs to a particular version.
Release records bind the experiment and strategy code to the reviewed evidence and intended execution stage. A changed strategy needs a matching release; paper admission does not authorize real-money execution.
Questions, answered.
More questions and answersAre all thresholds automatic hard blocks?
No. The implementation distinguishes hard validity checks, reviewable quality criteria and advisory findings. Some new criteria are introduced in advisory mode.
Does release approval remove investment risk?
No. A recorded review is a process control, not a promise of returns or a guarantee against loss.
What does the independent risk reviewer examine?
The risk reviewer examines strategy code, backtest and optimization reports, QA findings and measured promotion criteria. It reviews proposed exceptions separately from the analyst’s argument and can reject them. Read both reviews to understand where they agree or differ.
Does an analyst’s exception automatically approve a strategy?
No. A reviewable quality exception needs a recorded justification and risk review. An enforced hard failure cannot be waived by the analyst. A favorable review also does not by itself establish paper eligibility; the admission and release checks still apply.
Does approval carry over when the strategy changes?
Approval is tied to the reviewed experiment and strategy code. Release checks compare that identity with the run being admitted. A new version needs a matching release record; an older approval is not permission to execute changed code.
How should I interpret an advisory or unevaluated finding?
An advisory finding is reported for consideration rather than treated as an enforced failure. An unevaluated item means the required metric was unavailable. Neither label proves that the underlying risk is absent. Read the recorded values, limitations and review explanation.
Start with a question worth testing.
Explore published strategies, or open a workspace to build, test and monitor your own with AI agents.
Stratmill is an AI trading strategy platform, not financial advice or a broker. Backtest and paper results are hypothetical. Trading involves risk of loss.