Add agent × model comparison view with token-aware metrics - #4
Merged
Conversation
- Persist benchmark_key, task_key, model_id, review_mode, and judge_total_tokens per run (model was previously dropped on insert, making model-vs-model comparison impossible from history) - Add a Comparison workbench view and an 'agent-bench matrix' CLI command: setups side by side with score, quality, tests, duration, agent/judge tokens, and cost, plus a per-task score heatmap — aggregated from persisted runs at zero token cost - Add a zero-token review mode (UI toggle, --no-judge, or AGENT_BENCH_REVIEW=off) that skips the LLM judge while keeping all deterministic metrics - Surface free reviews (cache hits, rules/off) vs fresh gateway calls so token spend per comparison is visible - Cover run metadata persistence, matrix aggregation, and judge-off in unit tests; extend the e2e flow with the Comparison tab Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MZ5iDN2ezGTBFMs43SJ91N
…parison - Sandboxed runs producing an index.html now capture page.png (full-page screenshot) and page.webm (bounded scroll-through video) via local Playwright rendering — best-effort, zero model tokens, opt out with AGENT_BENCH_VISUALS / AGENT_BENCH_VISUAL_VIDEO - Persist screenshot_file/video_file per run; artifact route serves image/png and video/webm - Comparison tab gains a per-task Visual comparison gallery: latest visual run of each agent × model setup side by side, screenshot as video poster, click-to-play; run inspector shows the same media - Multimodal screenshot judge reuses the captured page.png instead of rendering twice - Tests: capture toggles, no-index.html fallback, real screenshot+video capture (skips cleanly when no Chromium is available) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MZ5iDN2ezGTBFMs43SJ91N
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
2 participants
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.
judge_total_tokens per run (model was previously dropped on insert,
making model-vs-model comparison impossible from history)
command: setups side by side with score, quality, tests, duration,
agent/judge tokens, and cost, plus a per-task score heatmap —
aggregated from persisted runs at zero token cost
AGENT_BENCH_REVIEW=off) that skips the LLM judge while keeping all
deterministic metrics
so token spend per comparison is visible
unit tests; extend the e2e flow with the Comparison tab
Co-Authored-By: Claude Fable 5 noreply@anthropic.com
Claude-Session: https://claude.ai/code/session_01MZ5iDN2ezGTBFMs43SJ91N