A live system is not tested by looking at it. It is tested by running it and writing down what it decided before the outcome existed. The log is the only artefact that makes a claim falsifiable.
About 1 min of reading
Write one decision before its outcome
Log the decision with stamps
Before you run it, say out loud what it will print. Then open the answer.
Record the outcome later
Before you run it, say out loud what it will print. Then open the answer.
Prove the decision came first
Before you run it, say out loud what it will print. Then open the answer.
Paper costs nothing and measures the same decisions. The only reason to skip it is a reason that costs money.
Long enough that the cost is visible. A cost that is invisible at ten trades is visible at two hundred, and that is the minimum.
A log written after the fact is a story. The timestamp of the decision is what makes it a record.
Model version, feed version, and size. Three stamps, and any past decision can be replayed from them.
The log has one reader: the person deciding the next change. If nobody reads it, the system is unmeasured.