"Improved" cannot be checked. A named state can be checked, which is the whole point.
Restate the claim as a specific condition with a value or threshold.
Evidence without dates cannot establish that the change followed the intervention.
Attach the underlying record and its date to each claim in the pack.
Self-reported evidence is the weakest form and the first thing a sceptic attacks.
Find a second source for each material claim, or mark it as self-reported.
Measurement designed after the outcome tends to measure the outcome you got.
Show the measurement definition with a date preceding the result period.
Changing the method mid-stream produces improvement that is entirely methodological.
Document both measurement methods side by side and reconcile any difference.
Measuring a clean subset and claiming the whole is the most common quiet overstatement.
State the measured population and the claimed population explicitly. If they differ, say so.
Unnamed exclusions are indistinguishable from cherry-picking, however innocent the reason.
List every exclusion with its rationale and its size.
Limits discovered by the audience are far more damaging than limits you declared.
Add a one-line scope limit to the claim itself, not a footnote.
If the record cannot be found, the claim rests on trust, and trust is not proof.
Include a reference path to the source record for every claim.
An assessment tells you how wide. Foundation teaches your people to close them without buying the read again.