feat: eval baseline arm — measure LCD vs bare Claude Code - #12
Merged
Merged
Conversation
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VRajFkyg9aBwbUFMhVaxNe
Red-first per the Deep-lane pipeline (work-item t1-baseline-eval in the dev harness). Covers: --baseline dry-run + symmetric isolation contract (AC-1), workspace strip / fail-closed residue + flag contradiction (AC-2), grader baseline row shape + two-way fail-closed (AC-3), cross-arm row comparability + frozen-prompt feature-paragraph parity (AC-4). AC-5 is a harness-closeout criterion, verified at audit/closeout, not unit-testable here. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VRajFkyg9aBwbUFMhVaxNe
…olation in run-eval.sh Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VRajFkyg9aBwbUFMhVaxNe
…aseline, two-way fail-closed Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VRajFkyg9aBwbUFMhVaxNe
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VRajFkyg9aBwbUFMhVaxNe
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
T1 of the 2026-08 product review (dev-harness
PROPOSALS-2026-08.md, work-itemt1-baseline-eval, decision D-025).Every results row so far compares LCD variants against each other; the headline claim — context economy pays against a plain agent — has never been measured. This PR adds the missing arm.
What
evals/run-eval.sh --baseline: bare-agent arm — no plugin, workspace stripped of the fixture's LCD onboarding (strip verified fail-closed), driven by a new frozen prompt whose feature paragraph is byte-identical to the LCD prompt's (test-enforced),EVAL-DONEfinish line.--baseline+--plugin-dirrefused as a contradiction.evals/grade.sh --baseline: identical row shape withaudit: n/a; fail-closed both directions; golden-locked criteria untouched (golden test passes unedited).~/.claude/CLAUDE.mdleak intoclaude -pby default, and--plugin-diradds rather than replaces — every model call now runs with a freshCLAUDE_CONFIG_DIR+--settings '{"enabledPlugins":[]}'. Arms differ by exactly one flag.tests/test-eval-baseline.sh: red-first fixture tests for AC-1..AC-4 (own file, so the old eval-harness work-item's identical AC literals can't green this audit).Notes
🤖 Generated with Claude Code
https://claude.ai/code/session_01VRajFkyg9aBwbUFMhVaxNe