# Release 2026-q4.1: diff vs 2026-q4

## Headline counts

| | |
|---|---|
| Occupations added | 0 |
| Occupations removed | 0 |
| Slug changes (break inbound links) | 0 |
| Tasks added / removed / text-edited | 0 / 0 / 0 |
| Employment figures changed | 0 |
| Pay figures changed | 0 |
| Occupation scores moved | 0 |
| Prompt changed | no |
| Model changed | no |
| Cache hit rate | 63.3% (34,020 hits, 19,690 misses) |
| Estimated scoring cost this refresh | not measured |

## Stability gates

✅ No drift flag. Diff is within the pre-registered thresholds.

### M-9 canaries (spec/04 §1.6.4)

| Canary | Occupation | OccScore | Expected | |
|---|---|---|---|---|
| Paralegals and Legal Assistants | `us-soc-23-2011` | 47 | 37–67 | ✅ |
| Paralegals and Legal Assistants | `uk-soc-3520` | 56 | 37–67 | ✅ |
| Heavy and Tractor-Trailer Truck Drivers | `us-soc-53-3032` | 18 | 13–31 | ✅ |
| Heavy and Tractor-Trailer Truck Drivers | `uk-soc-8211` | 20 | 13–31 | ✅ |
| Registered Nurses | `us-soc-29-1141` | 29 | 17–53 | ✅ |
| Registered Nurses | `uk-soc-2231` | 24 | 17–53 | ✅ |

## What a journalist would notice

Nothing changed. Sources hashed identical to the previous release, so every task was a cache hit and the emitted data is byte-identical. The runbook default is **not to promote**. The changelog gets a "verified current against sources" line instead, and the site's "last verified" date moves. Refresh cost: not measured of model spend, at a 63.3% cache hit rate.