gilde

← All craftbooks

Content Accuracy Review

Review and QAevalv1.0.0released 2026-06-05workflow: build-loop

Fact-check a piece of content for accuracy and produce a claim-by-claim verification report. First scopes the content and extracts every checkable claim (statistics, dates, quotes, attributions, causal and factual statements) and locks a verification standard, then verifies each claim against authoritative sources rating it true/false/misleading/unverifiable with the supporting evidence, then writes a report flagging every inaccuracy with the correction and source plus an overall trust rating. Use this for a fact-check, accuracy review, content verification, checking claims and statistics, source-checking an article, or vetting content before publishing — it returns a claim-level verdict backed by cited sources rather than a vague 'seems accurate'.

Steps

Entry step: scope. Each step names the specialist role it wants; the full working prompt is expandable.

  1. Scope + extract claimsreviewerentry

    extract every checkable claim, lock the standard

    Show working prompt
    Step 1: Read the content and EXTRACT every checkable claim into a numbered list — statistics and numbers, dates, named quotes and their attributions, causal statements ('X causes Y'), definitions, and any assertion presented as fact. Step 2: For each claim, note what kind it is and what evidence would settle it. Step 3: Separate verifiable facts from opinion/analysis (opinions are out of scope for accuracy but note unsupported opinions stated as fact). Step 4: Write the acceptance-criteria checklist / verification standard, e.g. 'every factual claim is extracted', 'each claim is checked against at least one authoritative source', 'a claim with no reliable source is marked unverifiable, not assumed true', 'corrections cite their source'. Step 5: Note the source quality bar (primary/official sources over secondary). Write the numbered claim list + standard to `write_task_note` AND `notes/scope.md`. No verification yet.
  2. Verify each claimresearcher

    check claims against sources, rate each

    Show working prompt
    Step 1: For EACH numbered claim, find authoritative evidence (use web search/fetch if available, prioritizing primary and official sources; otherwise reason from well-established knowledge and clearly mark anything you cannot externally confirm). Step 2: Rate each claim: TRUE (supported), FALSE (contradicted), MISLEADING (technically true but framed to deceive, or missing critical context), or UNVERIFIABLE (no reliable source found). Step 3: For every claim capture: the rating, the supporting source(s) with title/URL, a one-line evidence note, and — for false/misleading — the CORRECT statement. Step 4: Be adversarial: check whether a statistic is current, whether a quote is accurate and in context, and whether an attribution is right. Step 5: Do NOT mark a claim true just because it is plausible — require evidence, and prefer 'unverifiable' to a guess. Good looks like: each claim with a verdict and a citable source. Stage the verified claim table to `write_task_note`. Do not write the report yet.
  3. Write the accuracy reportreviewer

    claim-by-claim verdicts + corrections

    Show working prompt
    Step 1: Open `accuracy-review.md` and write an overall trust rating (e.g. Accurate / Mostly accurate with fixes / Significant inaccuracies) with the count of false/misleading/unverifiable claims. Step 2: Add an '## Issues' section listing every FALSE, MISLEADING, and UNVERIFIABLE claim first: the original claim, the verdict, the correct statement, and the cited source. Step 3: Add a '## Verified claims' section confirming the claims that checked out (with sources), so the review's rigor is visible. Step 4: Add a '## Sources' list of everything consulted. Step 5: Note any claim that needs a subject-matter expert. Every issue must include a correction and a source citation; never assert a correction without evidence. On a loop-back, fix only the named gaps. `write_task_note` the report path and the counts by verdict.
  4. Evaluatereviewer

    Grade the deliverable against every acceptance criterion. All pass → finish; any fail → loop back and fix the gap.

    Show working prompt
    Open `accuracy-review.md` and grade it against the scope standard. Check EACH criterion and write PASS/FAIL with a one-line reason: (1) every checkable claim from the content was extracted and addressed; (2) each claim has a verdict (true/false/misleading/unverifiable); (3) each false/misleading claim has a correction AND a cited source; (4) verified claims cite their sources; (5) unverifiable claims are marked as such rather than assumed true; (6) a sources list is present; (7) the overall trust rating matches the findings. If any criterion fails, name the exact gap and loop back.

Then route — this is the whole point of the loop:

- **Every criterion PASSES →** call `advance_task_step({ ref, stepId: "evaluate", next: "finish" })`.
- **Any criterion FAILS →** write the specific gaps to notes, then call `advance_task_step({ ref, stepId: "evaluate", next: "report" })` to loop back. The builder fixes exactly those gaps.

Never route to `finish` while any criterion is unmet. The build phase's completion gate already blocked a grossly-incomplete deliverable; your job is the judgment an automated check cannot make (does it actually work, read well, look right). After ~3 unproductive loops, stop and report DONE_WITH_CONCERNS so the user can step in.
  5. Finishdeveloper

    All acceptance criteria met. Stamp a short summary and report DONE.

    Show working prompt
    Every acceptance criterion passed. Write a one-paragraph DONE summary to task notes via `write_task_note`: what was built, the deliverable path(s), and a one-line confirmation that each criterion is met. Then report DONE.

Triggers

Phrases that suggest this craftbook to a crew.

Source

View this craftbook on GitHub · MIT license