Decision Evidence Brief
Answer a specific decision question with a weighted, confidence-rated evidence brief: a planner frames the decision and the options, locks the evidence criteria and a confidence scale, a researcher gathers evidence for and against each option and rates the strength of each piece, and a copywriter writes a decision-oriented brief that states a recommendation with explicit confidence levels and the key uncertainties. The ordering matters because evidence gathered before framing the decision is unfocused and a recommendation without a confidence scale hides how strong the case really is; fixing the question, the options, and a confidence rubric first produces an honest, decision-ready brief. Use this for evidence-based decision support, 'what does the evidence say we should do about X', recommendation memos, weighing options with confidence levels, and policy/strategy briefs grounded in sources.
Steps
Entry step: scope. Each step names the specialist role it wants; the full working prompt is expandable.
- Frame the decisionplannerentry
decision question, options, evidence + confidence rubric
Show working prompt
1. State the exact decision question and the discrete options on the table (including 'do nothing' if relevant) in a few lines. 2. Lock the criteria that good evidence must meet for this decision (relevance, source quality, recency) and a confidence scale: High / Medium / Low, with a one-line definition of each (e.g. High = multiple independent strong sources agree). 3. Decide how evidence maps to a recommendation (e.g. weigh strength-rated evidence for/against each option). 4. Write an acceptance-criteria checklist (e.g. 'evidence is gathered for AND against each option', 'each piece of evidence is rated for strength and cited', 'the recommendation states an explicit confidence level', 'key uncertainties and what would change the recommendation are listed', 'no recommendation exceeds the confidence its evidence supports'). 5. Record decisions. Call write_task_note with the question, options, confidence scale, and checklist; write the same to notes/scope.md.
- Gather & rate evidenceresearcher
evidence for/against each option, strength-rated + cited
Show working prompt
1. For each option, gather evidence both FOR and AGAINST into notes/evidence.md: the finding, its source [n], the date, and a strength rating per the locked confidence scale with a reason (sample size, source quality, corroboration). 2. Actively seek disconfirming evidence for the option you suspect is best — steelman the alternatives so the brief is not a foregone conclusion. 3. Corroborate strong claims across independent sources before rating them High. 4. Explicitly record what is unknown and what evidence would resolve it. 5. Mark anything you cannot verify UNVERIFIED. Good looks like a balanced, strength-rated evidence file per option. Do not write the recommendation yet.
- Write the briefcopywriter
recommendation with confidence + key uncertainties
Show working prompt
1. Write brief.md: a bottom-line recommendation up top stating which option and at what confidence level (High/Medium/Low), then a section per option laying out evidence for and against with each piece's strength rating and citation [n], then a 'Key uncertainties & what would change this' section. 2. Calibrate the stated confidence to the evidence — do not claim High confidence on Low/Medium evidence; if the evidence is thin, say so plainly. 3. Present the alternatives fairly, including the strongest case against your recommendation. 4. Add a References list from notes/evidence.md so every [n] resolves. 5. Keep it decision-ready and scannable. 6. On a loop-back, fix only the named gaps. 7. Call write_task_note with the brief path and which criteria now pass.
- Evaluatereviewer
Grade the deliverable against every acceptance criterion. All pass → finish; any fail → loop back and fix the gap.
Show working prompt
Open brief.md and check EACH criterion PASS/FAIL: (1) evidence is presented for AND against each option (not one-sided); (2) each piece of evidence is strength-rated and cited (spot-check 3 against notes/evidence.md); (3) the recommendation states an explicit confidence level; (4) the stated confidence is calibrated to the evidence — a High-confidence call must rest on multiple strong corroborated sources; (5) a 'key uncertainties / what would change this' section exists; (6) the strongest counter-case is acknowledged. A one-sided brief, or one claiming more confidence than its evidence supports, FAILS. 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: "write" })` 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. - 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.
- evidence brief
- what does the evidence say we should do
- decision support memo
- recommendation with confidence levels
- weigh the options with evidence
Source
View this craftbook on GitHub · MIT license