Eval — ADRs
Architecture decisions in the eval field.
Decisions about batch evaluation: how eval reuses the canvas execution engine (graph-driven, not bespoke pipelines), worker-pool concurrency for env-side throughput, and the batched-inference tier for GPU-bound policy / LLM nodes.
| ID | Title | Date | Status | Supersedes |
|---|---|---|---|---|
adr-eval-001 |
Graph-driven batch evaluation via LoopRunner | 2026-04-07 |
accepted | — |
adr-eval-002 |
Worker-pool batch env + batched inference tier | 2026-04-17 |
accepted | — |
adr-eval-003 |
Subprocess-per-run job scheduler with cross-session admission | 2026-05-07 |
accepted | — |
adr-eval-004 |
Per-episode log storage and replay-parser ABC simplification | 2026-05-15 |
accepted | — |