01
Define the decision population first.
State which decisions are included, the period covered, the unit of analysis and material exclusions. Changing the population after seeing the result can manufacture an apparent effect.
02
Use an outcome and window that fit the decision.
An operational signal may mature in days or weeks while the financial result takes months. Record both, with the relevant observation window for each.
03
Make the comparison reproducible.
Retain the baseline, matching rule, calculation version and missing-data treatment. If no credible control exists, report the observed outcome and the absence of a control rather than claiming attribution.
04
Confidence belongs in the record.
Evidence coverage, selection, seasonality and concurrent policy changes affect what the result can support. Confidence should describe the evidence, not enthusiasm for the intervention.