# Sara's EMEA CSAR Loop — Chapter 8 Worked Example

> The CSAR-loop trace behind Sara’s Tuesday-morning «one-page summary of why Q3 missed plan in EMEA» request in Chapter 8 of *The Defensible Decision*. Two attempts: cold first, then Crystallize-Scope-Assemble-Refine.

**Wholly synthetic.** Any resemblance to a specific real regional-analyst CSAR session is coincidental; the two-attempt structure and the six-question Crystallize pattern are drawn from typical practitioner-observed behaviour.

## Files

- [`sara-emea-csar.csv`](sara-emea-csar.csv) — flat rows: attempt / step / label / detail
- [`sara-emea-csar.json`](sara-emea-csar.json) — canonical scenario contract with source hash

## What it represents

A composite trace of the CSAR loop on a real analyst ask. Attempt 1 is Sara typing the director’s question verbatim into Copilot; the four-page seven-visual output looks impressive and nobody trusts it. Attempt 2 runs the Crystallize move (ask Copilot what IT would need to know) and produces six clarifying questions that Sara answers in a few minutes. Copilot restates the corrected scope; Assemble produces a one-page draft; Refine catches an accessibility issue on brand palette; Sara ships in twelve minutes.

## Invariants (locked)

The scenario carries executable invariants that Chapter 8’s release contract enforces:

- Attempt 1 (cold): 4 pages, 7 visuals, `director_ready: false`.
- Attempt 2 CSAR clarifying questions: exactly 6, ids 1..6, no gaps.
- Every clarifying question has both a question text and a `sara_answer`.
- Scope restatement accepted.
- Assemble output is 1 page (down from 4).
- Time-to-ship under 30 minutes.
- CSAR compression: pages 4→1, visuals 7→3.
- Refine catches at least one accessibility or contrast issue.
- Director response contains «thanks» or «exactly» (the satisfied acknowledgement).

## Why the invariants matter

Chapter 8’s pedagogical point is that the CSAR loop is cheaper than cold prompting even when it feels like extra work. If the compression ratio collapsed (both attempts landed at similar page/visual counts), the loop would look decorative. If Sara’s answers were vague or missing, the loop would look ceremonial rather than load-bearing. The invariants keep the trace defensible against skeptical readers.

## What it does NOT represent

- A specific real EMEA Q3 miss (the underlying data is out of scope; the scenario is about the loop shape).
- Copilot’s specific phrasing (medium-independent; exact wording varies with the model version).
- Other CSAR primitives at the same depth (Scope, Assemble, and Refine are recorded in shape only; Crystallize is where the six-question detail lives because Crystallize is the load-bearing move for this scenario).

## Reproducibility

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

## License

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