# Local scenario evaluation

Dataset: synthetic-v1-seed-7.

20 sequential runs per scenario including persistence; p95 is nearest-rank.

400 products; 8 locations; 3,200 inventory observations. SQLite + factual templates.

| Scenario | Checks | p50 ms | p95 ms |
|---|---|---:|---:|
| standard | PASS | 3.753 | 3.864 |
| nearby | PASS | 3.284 | 5.25 |
| quantity | PASS | 3.387 | 5.352 |
| stale | PASS | 3.311 | 3.528 |
| no_match | PASS | 1.16 | 1.291 |
| upgrade | PASS | 3.714 | 3.842 |

## Measurement limits

Local synthetic workload only. Snowflake, Azure, cloud models and production outcomes are not measured.

Latency includes query, lookups, explanation and state persistence. No production SLA is claimed.

A passing scenario check is not an independent verification of the original client engagement.