Work a full example

The project is to produce a three-source evidence table and a two-page synthesis that preserves uncertainty and study limitations. The user is a student writing a balanced evidence brief. Definition of done: the intended action is clear, the candidate uses approved evidence, blocking safety checks pass, and another person can reproduce the key result.

Stage 1: prepare

Create the job card and collect a bounded question, date range, source inclusion rules, search terms, labeled excerpts, and citation metadata. Remove or replace prohibited material: share only excerpts you are permitted to process; remove participant identifiers, unpublished interview material, paywalled full text, and confidential research data. Add one ordinary case, one boundary case, and one hostile or misleading case. Record unknowns instead of filling them with plausible guesses.

Stage 2: draft

Topic: effects of later secondary-school start times on adolescent sleep and attendance. Propose one focused question, five search phrases including a skeptical angle, and inclusion/exclusion rules. Do not provide citations. Mark every factual statement that will require a source.

The first candidate should be A question bounded by population, intervention, outcomes, and time; search phrases include counterevidence and implementation costs; no fabricated bibliography appears. In this worked run, imagine it also exhibits one realistic defect from this set: invented citations; treating search snippets as evidence; collapsing correlation into causation; hiding disagreement; quoting a summary instead of the source. Do not hide the defect. Mark the exact criterion it violates and decide whether the cause belongs to context, instruction, model capability, or the surrounding process.

Stage 3: repair narrowly

Issue a targeted revision:

Revise only the failed criterion identified below.
Preserve all verified content and the original output contract.
Do not add facts or assets.
Return the corrected artifact plus a one-line change note.
Failed criterion: [paste criterion and evidence]

A narrow repair keeps the review surface understandable. If the model cannot repair without new authoritative information, pause and obtain that information.

Stage 4: verify and release

Now open every source, check author/date/method, follow quotes to the original, compare primary findings with limitations, and label unsupported claims. Record pass/fail evidence for each criterion and have the named reviewer make the release decision. AI can map a search and organize supplied evidence; it is not the evidence. The non-negotiable habit is opening and judging the original source. Save limitations in language the audience can understand.

The durable deliverable is not only the final result. It is a research trail with question, query log, inclusion rules, source cards, claim-evidence matrix, and citation audit. Write what surprised you, which check found it, what you changed, and which control should become the default. A clean retrospective distinguishes a prompt improvement from a data, tool, or policy change.