[Feedback] :thumbsdown: negative — recall_learnings for scope WAU email Loop HubSpot ClickHouse retention

Nobody has claimed this yet.

## Feedback (negative)

**Question:** recall_learnings for scope WAU email Loop HubSpot ClickHouse retention
**Tool:** `route_question`

### Corrected answer
recall_learnings semantic search does not reliably retrieve learnings using broad multi-keyword queries. Specific single-topic queries (e.g. 'scope WAU') return results; broad queries combining multiple topics (e.g. 'scope WAU email Loop HubSpot ClickHouse retention') return nothing. The search appears to match on narrow keyword overlap rather than semantic similarity. To reliably retrieve learnings, use specific single-topic queries rather than compound queries.

### Notes
At the end of the dogfooding session, a broad recall_learnings query returned zero results despite 15 feedback entries having just been submitted. A narrower query on 'scope WAU' returned the correct learnings. This suggests the recall_learnings embedding/search is not working well for multi-concept queries. The tool should either improve semantic matching or the documentation should advise using specific narrow queries. This is a discoverability issue — learnings stored under specific topics won't surface when users ask a broader question that spans multiple topics.

---
_Feedback ID: `5f5fdb06-5eda-46c2-a512-25ed368b2102` | Recorded at 2026-03-11T00:53:36.288Z_

No files or tests are named. Start by tracing route_question into recall_learnings and examining how broad multi-topic queries are searched; reproduce the contrast between “scope WAU” and the compound query. Done means broad queries retrieve relevant learnings, or the tool documentation clearly advises narrow single-topic queries.

Written by the indexing model from the issue text.