eval_base is the reproducible research base derived from hybrid; it does not replace it. The canonical research plan is the 2026-09-08 EQ1–EQ4 plan; doc 47 is historical. Historical auxiliary documents remain source provenance; current implementation status is defined here.
- Overview and old/new experiment mapping
- Workload methodology and provenance
- Aggregation and schema boundaries
- Qwen capacity definitions and metrics
- Coverage and fail-closed behavior
- Thermal coverage and integration assessment
- Canonical simulator/tool entry points
- Build setup and mechanism validation
- Integration manifest, branches and test evidence
Offline tools are optional and off by default. Runtime prefetch and thermal reliability are deferred. GPU baseline parity: NOT VERIFIED. Old experiments must retain their original source hashes; none become new measurements by being documented here.
Current campaign source is eval/eq1-eq4-implementation@254d65a66279fbaffc5c185d04fbe41dc8dbba44 plus the recorded document patch. See current state, execution and matrix. Device-only thermal history remains separate from current runtime integration. Old status below is not a live device check.