# Intelligence-Selection Record

Complete one record per consequential decision route.

## Decision route

- Decision being made:
- Inputs and evidence:
- Required output:
- Consequence of error:
- Maximum tolerable error or effect:
- Latency and availability need:
- Cost ceiling:
- Human authority and fallback:

## Candidate comparison

| Mechanism | Evidence fit | Quality | Explainability | Latency | Full cost | Failure detectability | Recovery | Authority fit |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Deterministic code |  |  |  |  |  |  |  |  |
| Optimization |  |  |  |  |  |  |  |  |
| Classical ML |  |  |  |  |  |  |  |  |
| Retrieval |  |  |  |  |  |  |  |  |
| Foundation-model call |  |  |  |  |  |  |  |  |
| Bounded agent workflow |  |  |  |  |  |  |  |  |
| Human review |  |  |  |  |  |  |  |  |

## Selection

- Smallest sufficient mechanism or hybrid route:
- Why simpler options fail the stated requirements:
- Why more complex options do not earn their added cost or risk:
- Software-enforced boundaries:
- Human review capacity and authority:
- Fallback behavior:
- Evaluation evidence required:

## Change and retirement

- Condition that invalidates this choice:
- Signal that a simpler mechanism now suffices:
- Cost, error, adoption, or policy retirement trigger:
- Review owner and date: