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.
来源信息
- 仓库
- ruvnet/agentdb
- 最近来源活动
- 2026年5月6日 15:22
- 检测到的 SKILL.md 语言
- 英语
- 星标
- 94
- Forks
- 16
安装方式
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
检查来源文件
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
正在显示 SKILL.md
SKILL.md
来源说明 · 只读预览- 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.
在 GitHub 查看