01 / Ask a bounded question
Start with something we can check.
Define the security, period, calculation, or software behavior under study. State the assumptions before interpreting a result.
Our approach
Make the result
possible to question.
A useful research note shows what was checked, what happened, and where the conclusion stops.
02 / Establish the inputs
Check what the data describes.
Keep source identifiers and dates. Compare conflicting records with evidence about the issuer. An unresolved mismatch can be a reason to stop the affected calculation.
03 / Describe the comparison
Show what the test measures.
Name the baseline, workload, units, and conditions. Keep a software check separate from evidence of investment performance or model capability.
04 / Preserve the limits
Publish the unanswered questions too.
Distinguish measured results, historical records, and proposals. Link sources and state which private inputs are unavailable. A public summary alone may not be enough to reproduce an investigation.