# Sara's EMEA CSAR Loop — Schema

Flat CSV in `sara-emea-csar.csv`. Two logical row groups distinguished by `attempt`.

## Columns

| Column | Type | Description |
| --- | --- | --- |
| `attempt` | integer | `1` for the cold pass; `2` for the CSAR pass. |
| `step` | text | For attempt 1: `prompt`, `output`. For attempt 2: `crystallize`, `question_N`, `scope`, `assemble`, `refine`, `ship`. |
| `label` | text (quoted) | Short label describing the step (e.g. the question text for `question_N` rows, or "restatement accepted" for the scope row). |
| `detail` | text (quoted) | The detail associated with the step (Sara’s answer for question rows; page count for assemble; correction description for refine). |

The `label` and `detail` fields are CSV-safe quoted strings so commas and quotes inside descriptions do not break parsers.

## Row inventory

- 2 attempt-1 rows (`prompt` + `output`).
- 12 attempt-2 rows (`crystallize` + 6 question rows + `scope` + `assemble` + 1 refine row + `ship`).

Total: 14 rows.

## Deliberate design choices

Unlike numeric datasets (Priya, Nadia, Kenji), Sara’s dataset carries no measured quantities. Its purpose is to lock the *shape* of the CSAR loop as a reproducible sequence. Every row is a locked fact that keeps the print figure (Fig 8.3), the chapter narrative, and this dataset in lockstep.

The one designed asymmetry is that attempt 2 carries far more detail than attempt 1 — attempt 1 is deliberately short because its whole point is that the cold pass surfaces no defensible detail; attempt 2 is where the discipline lives.

## License

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