# Diego's Campaign-Attribution Report — Chapter 9 Worked Example

> The cold-prompt vs CSAR-directed trace behind Diego’s Q3 campaign-attribution report in Chapter 9 of *The Defensible Decision*. Three failure modes the cold pass ships silently; three CSAR moves catch them.

**Wholly synthetic.** Any resemblance to a specific real marketing-attribution report is coincidental; the three-failure-mode structure is drawn from typical practitioner-observed patterns.

## Files

- [`diego-campaign-attribution.csv`](diego-campaign-attribution.csv) — flat rows: cold-pass failure modes + CSAR clarifying questions + Refine corrections
- [`diego-campaign-attribution.json`](diego-campaign-attribution.json) — canonical scenario contract with source hash

## What it represents

Diego’s one-line cold prompt to Copilot returns a report that LOOKS defensible; Copilot silently picked a measure that answers a related question (attributed *revenue* not attributed *pipeline*), picked individual-campaign grain (obscuring the campaign-type story), and produced a Smart Narrative that names top campaigns by raw pipeline dollars (which favors longer-active campaigns). Diego runs the CSAR loop; six Crystallize questions surface all three failure modes; Refine catches three surface-level polish items; Diego reconciles against finance close and ships.

## Invariants (locked)

The scenario carries executable invariants that Chapter 9’s release contract enforces:

- Exactly 3 cold-pass failure modes: `measure`, `grain`, `narrative`.
- Each failure mode names `copilot_pick`, `diego_intended`, and `detection`.
- Measure failure pins «Attributed Revenue» (Copilot pick) vs «Attributed Pipeline» (intended).
- CSAR Crystallize returns exactly 6 clarifying questions, ids 1..6, no gaps.
- Questions cover measure, grain, comparison window, anchor, narrative shape, and per-week efficiency.
- Ship verdict is `ship` and reconciliation kind is `finance_close_report` with matching outcome.
- Defensibility bundle carries at least 6 items (matching Chapter 8’s bundle standard).
- Cold pass is NOT ship-ready; CSAR pass IS ship-ready; problems caught by CSAR equals the failure-mode count.

## Why the invariants matter

Chapter 9’s pedagogical point is that Copilot’s report-generation failure modes are silent by construction; the CSAR loop makes them visible. If the failure count changed, or if any CSAR question missed one of the three failure classes, the «each failure caught by a specific question» mapping would break. If reconciliation against finance close were skipped, the ship decision would lose its defensibility anchor.

## What it does NOT represent

- Actual Q3 campaign attribution data (the story is about the CSAR discipline, not the numbers).
- Individual campaign performance (the dataset carries the *type* of failure per grain, not per-campaign attribution).
- The finance close report itself (out of scope; the reconciliation move is a check, not a joined dataset).

## Reproducibility

The dataset ships from the book source at `AlexBooks/applied-ai/ai-directed-viz/data/diego-campaign-attribution.json` and is re-emitted here by `AlexBooks/applied-ai/ai-directed-viz/scripts/generate-diego-attribution.mjs --site-root <site-root>`.

## License

Released for teaching use alongside *The Defensible Decision*. Use freely.