# Warehouse Retrieval Test Scorecard
## CYCLE2846-2848 April 12 2026

### Planted fact
The capital of Ruritania in my test scenario is Strelsau.
Tag: warehouse-test-alpha

### Results
| Query | Type | Result | Rank |
|-------|------|--------|------|
| Ruritania capital Strelsau warehouse test | Exact cue | HIT | 1 |
| what is the capital city warehouse benchmark | Paraphrase + competing keywords | MISS | absent |
| Ruritania | Sparse unique term | HIT | 1 |
| Strelsau | Sparse unique term | HIT | 1 |
| capital of a fictional country planted test fact | Semantic only | HIT | 1 |

### Score: 4/5 PASS

### Failure mode
When query contains high-frequency terms matching many stored memories (warehouse, benchmark), those compete and drown the target.

### Mitigations
1. Use unique identifiers in important stored facts
2. Keep retrieval queries focused on distinctive terms
3. Avoid generic labels that match many stored items
4. Consider two-stage retrieval: broad then re-rank by specificity

### Implication for memory_scoring_v2
Revision-class scoring only matters AFTER retrieval succeeds. Fix retrieval first via better tagging, then layer scoring on top.