# Prompt 2 — Generate a Comparison Report

> Adapted from Chapter 9 (Diego worked example, campaign attribution). The CSAR-loop entry point for plan-vs-actual reports against a strict finance measure.
>
> **Verified against**: Power BI Copilot, June 2026 surface

## Crystallize

> "Build me a one-page executive report on Q3 campaign-attributed pipeline against the annual plan, by campaign type, with the takeaway focused on whether marketing-sourced pipeline is on track for the FY plan. Use the `[Attributed Pipeline]` measure (the strict definition the finance team uses for plan-tracking), not `[Influenced Pipeline]` or `[Sourced Pipeline]`. Quarterly grain. Entity grain is campaign type, not individual campaign."

The Crystallize move names **measure**, **grain**, **comparison**, and **takeaway** — all four. When any of the four is missing, the Refine column gets longer.

## Scope

Copilot restates the measure, grain, comparison, and takeaway. Confirm. The restatement may also name a constraint Copilot inferred but you did not state (the Smart Narrative paragraph will be limited to one sentence, for example). Accept or reject explicitly — silent acceptance defeats the purpose of the scope turn.

## Assemble

Expect one page; clustered bar of quarterly contribution by campaign type; card showing year-to-date number against plan; one-sentence Smart Narrative.

## Refine (corrections named by Part I principle)

- **Ch 4 (chart-family fit)**: If the clustered bar is sorted alphabetically, ask for descending order by most recent quarter's contribution. Ordering carries argument; alphabetical order argues nothing.
- **Ch 6 (color discipline)**: If the card is colored green even though the run-rate suggests the year-end will not close the gap, ask for a neutral color until the year is closed. Color is the strongest visual claim; do not let Copilot make a claim the data does not yet support.

## Ship decision

Ship when the year-to-date number reconciles to the finance attribution close. The reconciliation is non-negotiable for plan-tracking artifacts.

---

**When to use this prompt**: The audience needs a strict plan-vs-actual view backed by a single finance-blessed measure. The discipline of naming the measure explicitly (especially when adjacent measures share the same general name) is what prevents the wrong measure from quietly entering the report.

**Dataset**: [diego-campaign-attribution](/datasets/diego-campaign-attribution/) — download CSV + JSON to reproduce the Diego comparison-report pass end-to-end.