| name | satisfaction-scoring |
| description | The probabilistic satisfaction metric. Not pass/fail — a judgment about the fraction of users who would be satisfied with this output. |
The Metric
The satisfaction score answers this question:
"Of a population of users who wanted what the spec describes, what fraction
would be satisfied with this output?"
Express as N/10. This is a judgment about fitness for purpose, calibrated against
the spec's stated intent — not a code quality score, completeness checklist, or
binary pass/fail.
Scoring Guidelines
9-10/10 — All verification criteria satisfied, no unexpected behaviors.
7-8/10 — Core behavior correct. Minor gaps most users wouldn't encounter.
5-6/10 — Core behavior partially correct. Missing behaviors a meaningful fraction needs.
3-4/10 — Key requirements not met.
1-2/10 — Fundamental misalignment with the spec.
Required Reasoning
Always accompany the score with:
- Which criteria were fully satisfied
- Which criteria were partially satisfied or missed
- Why you placed the score where you did rather than one point higher or lower
Threshold
A score of 8/10 or higher moves to tasks/verified/.
A score of 7/10 or below moves to tasks/failed/ with a full failure report.
