agentdb-explainable-recall
Search with feature attributions — return WHY each match scored where it did. Use when debugging recall quality, auditing for bias, or explaining results to a user.
Datos de origen
- Repositorio
- ruvnet/agentdb
- Última actividad en el origen
- 6 de mayo de 2026 a las 15:22
- Idioma detectado de SKILL.md
- inglés
- Estrellas
- 89
- Forks
- 14
Opciones de instalación
De forma predeterminada está seleccionado el prompt que primero revisa el origen. Puedes cambiar a un comando directo o descargar una copia local.
Revisa los archivos de origen
Lee SKILL.md y los archivos complementarios que muestra SkillsMP antes de decidir si quieres instalarlo.
Mostrando SKILL.md
SKILL.md
Instrucciones de origen · Vista previa de solo lectura- name
- agentdb-explainable-recall
- description
- Search with feature attributions — return WHY each match scored where it did. Use when debugging recall quality, auditing for bias, or explaining results to a user.
# Explainable Recall
Standard search returns scores; explainable recall returns *features* — which dimensions of the embedding (or which keywords in hybrid search) drove the match.
## When to use
- Recall quality is off and you need to debug *why*
- Auditing for bias / fairness
- User asks "why this result?" — show the receipts
- Building UIs that surface match rationale
## API
```
agentdb_explainable_recall(
query: <embedding | string>
k: 5
features: 'embedding-dims' | 'bm25-tokens' | 'hybrid-both' | 'metadata'
)
Returns: [
{
id, score,
explanation: {
topDims?: [{ dim: 12, contribution: 0.18 }, ...],
topTokens?: [{ token: "jwt", contribution: 0.31 }, ...],
metadataMatch?: { topic: 'auth', project: 'api' }
}
},
...
]
```
## Use cases
| Use | Features setting |
|---|---|
| Debug an unexpected high-score | `embedding-dims` — see which dims spiked |
| Verify keyword fall-back works | `bm25-tokens` — see if exact terms were the driver |
| Confirm metadata filters fired | `metadata` — see which filter values matched |
| Build user-facing UI | `hybrid-both` — show both text-level + dim-level signals |
## Don't
- Don't pipe the full feature vector to the user — it's noisy. Pick top 3-5 contributing dims/tokens.
- Don't use explainable recall for normal searches. The feature extraction adds 20-50ms per query.
Ver en GitHub