Find the right response to a research problem

Begin with the problem that is holding up the research. The field guide gives a compact starting protocol; the ten-step protocol explains the wider sequence. Neither asks researchers to complete every template.

Before sharing source material: check whether institutional and ethics requirements, consent, third-party rights, confidentiality agreements and service-specific terms permit disclosure to the proposed AI system. If permission is uncertain, do not transfer the material. An authorised environment or wholly synthetic example may be an alternative. Summaries and paraphrases can also disclose restricted, identifying or sensitive information; reassess the content against the same rules. Removing names alone does not guarantee safe disclosure.

Research situation	Decision requiring judgement	First action and resource	Illustration or reference
Planning AI use with restricted material	May these sources enter this particular service?	Establish the permitted-data boundary in the Research Charter before transfer.	Public-release policy addresses the separate downstream disclosure question.
Considering a useful AI suggestion	Is this a research option or a finding?	Specify what to check; record a consequential adoption, rejection or alteration in the Decision Log as it happens.	Moving-benchmark question
Inspecting generated code, extracts or citations	What independent evidence exists?	Execute, inspect or evaluate the proposed object as the design requires; use the Evidence Ledger for claims that emerge.	Code is not evidence
Evaluating a strong-looking result	What credible observation or alternative reading could weaken it?	Undertake a specific check or interpretive challenge; document a consequential result in a Claim Challenge Record.	Method claim narrows
Interpreting interviews or historical sources	Does the proposed theme or reading respect context, contrary material and source limitations?	Reinspect extracts, selection and contextual records, or source provenance as appropriate to the research design.	Illustration and qualitative transfer
Revising a plausible result’s meaning	Does the design warrant the intended conclusion?	Record an altered, narrowed or withdrawn inference in the Inference Revision Record.	Wrong inference
Preparing an attribution statement	Which histories does the preserved trace distinguish?	Assess contemporaneous evidence separately from later recollection in the Reason-Provenance Record.	Retrospective storytelling
Rewriting a manuscript	Does the revision preserve warranted qualifications and the distinctive contribution?	Use a Claim-Preservation Contract; reopen it explicitly if new evidence changes the authorised claim.	Finished scholarly object
A paper becomes safer but less interesting after repeated AI review	Was novelty withdrawn by evidence, or displaced by repeated stylistic advice?	Stop the review loop; compare the proposed change with the original contribution and concrete counter-evidence. Retain defensible novelty.	Novelty erosion and salience substitution
Making project material public	Are permissions, evidential status and quotations suitable for release?	Apply the Public Release Screening Checklist.	Evidence-status rules

What constitutes a useful next step?

A generated result that was never checked requires evidence production, not more polished writing. A checked result with an over-extended claim requires an inferential revision. A defensible claim that has become unclear can be edited under its existing boundaries. A claim about who supplied an idea requires a distinct trace, which is best preserved when the decision occurs, even though attribution is assessed later.

For a compact stage-to-record reference, see the protocol crosswalk. For a single invented decision followed through the stages, use the constructed illustration.