The verdict
The verdict is the step in the pipeline where you find out whether a strategy actually holds up. It is step five of seven: the one that tells you whether the next step is worth taking. Getting it right requires care, so this page explains exactly how it works.
The same approach shows up in Market Making today, in the confidence label that tells you when there is not enough history to trust an estimate.
How a strategy is judged
Section titled “How a strategy is judged”Out-of-sample by default
Section titled “Out-of-sample by default”The strategy is fit on one slice of history and judged on a separate, later slice it never saw. Only the test slice is reported. An in-sample curve is never shown as the result.
No look-ahead
Section titled “No look-ahead”A signal at time T may only use data available at or before T. No future bars, no using a bar’s own close to enter on that same bar. This is enforced by construction.
Measured against random
Section titled “Measured against random”Beating “no trade” is easy. ClinkAI compares your strategy against many random strategies that trade the same amount, and reports where yours lands.
- The headline is edge confidence: the percentile versus random (for example, “ahead of 93% of random strategies that traded the same amount”).
- It is deliberately not phrased as “probability your edge is real,” because the method cannot produce a forward-looking certainty like that.
Verdict bands:
| Percentile vs random | Verdict |
|---|---|
| Below 90th | Probably noise |
| 90th to 97th | Weak or uncertain |
| Above 97th | Worth a small live test |
The backtest profit is a supporting detail, not the headline.
Real costs
Section titled “Real costs”Gross and net are both shown, where net is after fees, funding, and slippage. If the result dies after costs, that is the headline. When data has gaps, ClinkAI says so instead of filling them in.
Sample-size gate
Section titled “Sample-size gate”Below a minimum number of trades in the test window (currently 30), ClinkAI declines to give a verdict: “not enough trades to conclude anything.” Saying nothing beats guessing.
The forking penalty
Section titled “The forking penalty”Test many variants of an idea and one will look good by chance. ClinkAI accounts for how many you tried, so a result that only looks good because you forked twenty variants does not get promoted.
What a verdict means in practice
Section titled “What a verdict means in practice”A clear “no” is a real result, not a failure. It tells you the next step is not worth taking yet, and that is the information you needed. When the verdict is positive, it means the strategy cleared the bar on the test window — not that future results are guaranteed. Use it to decide whether a small live test is justified.