RMP.4:5.3 - A computation with a research-shaping AI suggestion

A researcher uses an AI service to suggest a preprocessing change for a model comparison. In this constructed case, the retained account identifies the supplied design, original data and code version, the relevant proposal and input context, and the researcher’s decision to revise and use the change. The actual run connects the changed code, environment and parameters to its output.

The service also suggests a nonexistent supporting article. The researcher rejects it after source checking and retains the actual source used. The false citation remains a rejected suggestion; it does not become evidence merely because it is preserved in the trace.

Later, the original service version is unavailable. A reader can inspect the retained suggestion, the human decision and the implemented change. They cannot infer that the same service response can be regenerated. A new computational run may test the retained implementation, but it is new work and does not establish replay of the original AI interaction.