# Asha's Escalation-Driver Confound Check — Schema

Flat CSV in `asha-escalation-drivers.csv`. One row per driver.

## Columns

| Column | Type | Description |
| --- | --- | --- |
| `id` | integer | Driver id (1..3) in the order Key Influencers ranked them. |
| `driver_name` | text (quoted) | Human-readable driver name as it appears in the Key Influencers UI. |
| `raw_effect_multiplier` | decimal | The multiplier the AI reports (e.g., 4.1 means «increases escalation by 4.1x»). |
| `verification_move` | text (quoted) | The specific check Asha ran on this driver (add co-Explain-by field, filter down to a segment, etc.). |
| `verified_verdict` | text | One of `survives`, `survives_when_scoped`, or `confound`. |
| `verified_notes` | text (quoted) | What the verification move revealed. |
| `action` | text (quoted) | The follow-up action (surface, do not surface, rescope, etc.); empty for clean survivors. |

## Row inventory

- 3 rows — one per driver.

Total: 3 rows.

## Deliberate design choices

The dataset is small on purpose: Chapter 10’s point is that three drivers on a Key Influencers page can produce three different truthfulness verdicts once verified, and the reader needs to see the whole space (clean survivor, scoped survivor, confound) in one glance.

Raw effect multipliers are stored to one decimal to match the Key Influencers UI convention. Verification moves are recorded as sentences (not codes) so the CSV doubles as a checklist the reader can copy directly.

## License

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