Retrain archetype brains: labels have drifted from their trained state #4111

The archetype brains have fallen out of step with the engine. Labels in DEGRADED levels are still emitted but qualified; INVALID levels are withheld (Unvalidated). This issue is opened, updated and closed by archetype-validation.yml (#4100) -- it closes itself once every level is back to VALIDATED or DRIFTING.

Current state

trained against engine v2.9.1 (44ffeb1, 2026-09-19); last measured at engine f81ab92 on language-crucible@v1.7.0

Eroding archetypes (share of trained-state members still assigned):

• function C Struct Operations: 95.1% retained
• function Encapsulated Accessors: 85.6% retained
• function I/O & Config Routines: 92.5% retained
• function State Mutators: 93.4% retained
• function Stateful Encapsulated Methods: 86.4% retained
• file file_cluster_13: 97.4% retained, mostly to 'file_cluster_15'
• file file_cluster_15: 97.9% retained, mostly to 'file_cluster_10'
• file file_cluster_16: 55.6% retained, mostly to 'file_cluster_15'
• file file_cluster_6: 94.5% retained, mostly to 'file_cluster_4'
• file file_cluster_8: 81.6% retained, mostly to 'file_cluster_6'
• file file_cluster_9: 92.3% retained, mostly to 'file_cluster_2'
• composition_file Callbacks & Closures Files: 95.3% retained, mostly to 'Large Core Modules (2)'
• composition_file Compute Cores Files: 95.4% retained, mostly to 'Large Core Modules (3)'
• composition_file Encapsulated Accessors Files: 83.0% retained, mostly to 'Large Core Modules (2)'
• composition_file Interface Declarations Files: 98.0% retained, mostly to 'Large Core Modules'
• composition_file Large Core Modules: 92.1% retained, mostly to 'Interface Declarations Files'
• composition_file Large Core Modules (3): 90.4% retained, mostly to 'Interface Declarations Files'
• composition_file Many-Argument Workhorses Files: 90.0% retained, mostly to 'Large Core Modules (2)'
• composition_file State Mutators Files: 93.5% retained, mostly to 'Parameter Forwarders Files'

Agreement = share of crucible code files labelled as at training; stability, not accuracy.

Clearing it

1. Retrain/refreeze the brains in gitgalaxy-population-analyses on a current fleet scan.
2. Land them here with the paired engine review (EXPECTED_CONTRACT_SHA in tests/core_engine/test_archetype_parity.py) and a golden-master re-bless.
3. Re-anchor the drift baseline on the engine the brains were trained against:
   python tests/tools/archetype_drift.py rebaseline --audit <crucible audit> --engine-version vX.Y.Z --engine-commit <sha> --trained-at <date> --corpus <corpus>

Recent measurements