 xiaoju
|
8c26f16716
|
feat: builtin judges — frontmatter + token-stats (deterministic) + upstream/hallucination (stubs)
CI / check (pull_request) Successful in 1m45s
Implement 4 builtin judges for eval framework:
- frontmatter-compliance: validates YAML frontmatter with $status field,
score = stepsValid / stepsTotal
- token-stats: aggregates Usage from step nodes, always score 1.0
(informational only)
- upstream-consumption: LLM-as-judge stub (score 0, TODO)
- hallucination: LLM-as-judge stub (score 0, TODO)
Infrastructure:
- judge/builtin/read-steps.ts — shell out to uwf step list
- judge/builtin/types.ts — BuiltinJudge, BuiltinJudgeOutput
- runner/collect.ts — dispatch builtin judges by name
9 new tests (frontmatter validation + token aggregation)
Refs #71
|
2026-06-05 00:09:06 +00:00 |
|
|
xiaoju
|
fae9e9ed3a
|
feat: eval run command — prepare, execute, collect pipeline
CI / check (pull_request) Successful in 1m45s
Implement the uwf-eval run <task-dir> command with 3-phase pipeline:
- prepare: read task.yaml, copy fixture/ to temp workdir
- execute: shell out to uwf thread start + exec
- collect: run judges, compute weighted score, store CAS node,
set @uwf/eval/<task>/latest variable
Changes:
- src/runner/ — types, prepare, execute, collect, index
- src/storage/store.ts — createEvalStore(), setEvalLatest()
- src/commands/run.ts — full pipeline wiring with --agent/--model/--count
- 9 new tests (prepare + collect + weighted scoring)
Builtin judges return placeholder score 0 (Phase 1c).
Refs #70
|
2026-06-04 23:59:21 +00:00 |
|