Library / Research Method Practice Principles Framework
Jump to passage
In this reading

Link to current text

Published source confirmed at last check

Source changed 2026-10-03 10:39:28 UTC · snapshot created 2026-10-03 10:40:04 UTC · last check 2026-10-03 11:35:10 UTC

RMP.4:4.4 - Explain consequential changes and contributions

Preserve the earlier intention and actual change wherever their difference matters. State what changed, when, why, who made the consequential choice, which material it affects and what relevant data or results were available by then. Record uncertainty instead of filling an unknown reason with a plausible one.

Distinguish events with different consequences. A missed observation, excluded case, failed run, inaccessible source and revised interpretation are not interchangeable “cleaning.” A record can be intact yet excluded from a particular analysis; an observation can be absent because the event was missed rather than because its value was extreme.

If a choice was made after relevant outcomes were seen, retain that timing. Such a choice may still be justified for the actual Method and use, but cannot supply an advance commitment. A later sensitivity analysis or independent criticism is separately performed work, not a retrospective repair of the original timing.

Identify substantial human and tool contributions. For AI-assisted work, preserve what is needed to inspect the research-shaping contribution:

  • the model or service identity available to the user, version or configuration when known, and timing;
  • the relevant input context and output actually used, subject to the applicable protection and access conditions; and
  • the responsible person’s checks, acceptance, rejection or revision and the effect on the research work.

Exact prompt content belongs in the retained trace when it can change the examined research choice. Routine spelling or editorial assistance need not generate a transcript archive. An unavailable internal model state remains unavailable; do not invent it to make a service appear replayable.

If an AI output proposes a source that does not exist, preserve its rejection where that matters to the account and use the actual checked source as evidence. The rejected output is not a source in the scientific chain.