Dataset to Narrative Report
Turn a dataset into a written, evidence-backed analytical report that a non-technical reader can act on. Analyzes the data FIRST to surface the real findings (key metrics, segments, trends, surprises) with supporting numbers, then writes a narrative that leads with the takeaway and traces every claim to a figure, then builds the polished report document with sections, a summary table, and an explicit limitations note. Use this for turning raw numbers into an insights report, executive summary, data storytelling, or a findings write-up stakeholders will read.
Steps
Entry step: analyze. Each step names the specialist role it wants; the full working prompt is expandable.
- Analyze the datadata-analystentry
extract the real findings with supporting numbers
Show working prompt
Interrogate the dataset to find what actually matters — do not write prose yet. Step 1: state the 3-5 questions the report should answer (totals, trends over time, top/bottom segments, notable correlations or anomalies). Step 2: compute the answer to each with concrete numbers (counts, means, %, deltas, top-N) and record the exact figure plus how you derived it. Step 3: identify the single most important takeaway and 2-3 supporting findings, each tied to a number. Step 4: note data caveats (sample size, missing periods, definitions). Step 5: write an acceptance-criteria checklist for the report ('leads with the headline finding', 'every claim cites a number', 'includes a summary table', 'states limitations'). Call write_task_note and write the findings + figures + checklist to notes/analyze.md. No narrative prose yet. - Write the narrativecopywriter
draft a takeaway-led narrative grounded in the figures
Show working prompt
Write the report narrative using ONLY the figures from the analysis note — invent no numbers. Step 1: open with a 2-3 sentence executive summary that states the headline finding and its 'so what'. Step 2: write one short section per supporting finding; each paragraph states the insight, then the number that proves it, then the implication. Step 3: keep it scannable — short paragraphs, plain language, no jargon a manager wouldn't use. Step 4: end with a clearly labeled 'Limitations' paragraph from the caveats. Call write_task_note with the draft keyed by section so the build phase drops it straight in.
- Build the reportdeveloper
assemble the polished report.md with sections and a summary table
Show working prompt
Assemble the final report as report.md. Step 1: add a title, a date, and an Executive Summary section from the approved narrative. Step 2: add one ## section per finding with the prose and inline figures. Step 3: include a Markdown summary table of the key metrics (metric, value, change). Step 4: add a Methodology / Data notes section and a Limitations section. Step 5: ensure every numeric claim in the prose matches the analysis note. On a loop-back, fix only the named gaps. Call write_task_note with the report path and section list.
- Evaluatereviewer
Grade the deliverable against every acceptance criterion. All pass → finish; any fail → loop back and fix the gap.
Show working prompt
Read report.md as a stakeholder would. Verify: it opens with a takeaway-led executive summary; every numeric claim matches a figure in the analysis note (re-check at least three); it contains a summary table of key metrics; section headers organize it; and a Limitations section is present and honest. Confirm no fabricated numbers and no claim without a supporting figure. Write PASS/FAIL per acceptance criterion; any FAIL loops back to build. 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: "build" })` 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.
- turn data into a report
- analyze this dataset
- write a findings report
- data insights writeup
- executive summary from data
Source
View this craftbook on GitHub · MIT license