RMP-CORRECTION — Repair an analysis and return its consequence without discarding a suitable method

• Situation: A completed analysis contains a calculation defect, and a later change in the observations may also challenge the method’s fit.
• Question: Which object needs correction now, and when does a different method comparison become necessary?
• First useful result or blocker: A verified implementation correction with the affected analysis and claim returned to their users; under changed receiving conditions, a supported method decision or the still-unresolved qualification.
• Start with: RMP.9 — Decide Whether and How to Improve Research Methods and Practice, using the actual difficulty and receiving conditions.
• Stop or return: Retain a suitable method after a determinate repair. A selected comparison under changed conditions remains a plan until its results exist.

1. Locate the defect and retain its scope. In the calculation application, the supplied method is suitable for the independent observations 2, 4, 6 and 8 and specifies sample variance with denominator n−1. Their mean is 5 and squared-deviation sum is 20. The implementation divides by n, reporting 5 instead of 20/3. RMP.4 identifies that computation and the analyses that used it.
2. Correct and verify the affected result. RMP.9 retains the method and repairs the denominator. RMP.5 recalculates the dependent analysis; RMP.8 returns any changed uncertainty or conclusion to the actual use that relied on it. The known-answer check settles this correction without new participants or an efficacy study. Unaffected analyses retain their existing support.
3. Reopen fit when the question changes. Now suppose the programme needs an equally weighted mean across independent sites, with unequal numbers of dependent observations within each site. The old pooled-observation analysis no longer answers that target under the case premises. RMP.9 compares candidates for the same site-level quantity. In the changed-situation case, a qualified statistician, existing simulation code and a two-day allocation make a useful comparison attainable: a t interval from independent site means versus resampling whole sites. The selected comparison is the result so far; the winning procedure and its qualification remain unresolved until actual results exist. Retain the restriction on the unsupported uncertainty claim and reconsider the acquisition if access, capability or worth changes.