feat(evals): T1-5.1 — re-baseline on claude-fable-5-1, persisted quality rubric, evaluator priced (0.19.0) - #19
Merged
Merged
Conversation
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01R6BjXKFWAvEtKDYifc2weC
…ault model claude-fable-5-1, persisted quality rubric - evals/fixture conventions: closeout-evaluator: on (its cost lands in the row) - run-eval.sh: default EVAL_CLAUDE_MODEL claude-fable-5-1 (rows stamp the id; D-015) - evals/rubric.md: the 5-dim two-lens rubric as an artifact; evals/quality/ for scoring files - tests: fixture asserts the key; runner dry-run asserts the default model Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01R6BjXKFWAvEtKDYifc2weC
…5 workspaces (18 judges, two lenses) Deep 68 / Standard 70 / bare 68 of 75 — the parity previously quoted from transcripts now has a file. Checksum-skips-a-record mutation survives in 5/9 workspaces across arms. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01R6BjXKFWAvEtKDYifc2weC
…ic scoring artifact, evaluator record; README numbers; CHANGELOG 0.19.0 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01R6BjXKFWAvEtKDYifc2weC
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Harness D-027 → ROADMAP T1-5.1. Every prior benchmark row ran
claude-fable-5; sessions now run 5.1 and D-015 forbids cross-model comparison, so the headline numbers were stranded. This PR re-measures all three arms on 5.1 and turns quality into an artifact.Changes
evals/fixtureconventions:closeout-evaluator: on— the opt-in evaluator's cost now lands in the plugin rows (fixture test asserts the key).evals/run-eval.sh: default modelclaude-fable-5-1(dry-run test pins it; rows stamp the id).evals/rubric.md+evals/quality/: the two-lens rubric as a persisted artifact, applied to the 9 new 5.1 workspaces and retroactively to the 9 verified fable-5 workspaces (18 independent judges each, none saw another's score).evals/results.md: 9 new rows + the T1-5.1 conclusion. README numbers updated. CHANGELOG 0.19.0.Results (n=3/arm, one fixture,
claude-fable-5-1)Parity across arms on both tiers (fable-5 on the same rubric: 68/70/68). Evaluator challenged 3/6 plugin closeouts, twice on a real cross-surface parity defect (
GET /stats?returned 200 while the CLI rejected the empty arg); all fixed before closeout.Tests 13/13,
plugin validatepasses, versions in lockstep (0.19.0,feat/→ minor).🤖 Generated with Claude Code
https://claude.ai/code/session_01R6BjXKFWAvEtKDYifc2weC