docs(benchmarks): record the 20-task terminal-bench pilot baseline - #34
Merged
Conversation
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Document a reproducible 20-task Terminal-Bench 2.1 run with pinned agent and dataset revisions, batch commands, cost scenarios, and a plan for the full 89-task baseline. Preserve compact machine-readable results for future harness comparisons and regenerate the full English development notes from the Chinese source.
The run scored 8/20 (40%) by Harbor reward and 7/20 (35%) without a Harbor exception. Its 267 recorded model replies cost $0.119200332 after supplemental reconciliation; wall time was 111 minutes 10 seconds. The notes document 11 failures ending at
max_tokens, one exhausted turn budget, and continued agent execution during verification after a timeout. These observations inform subsequent fixes and a repeat of the same 20 tasks.The latest result file retains experiment configuration, task hashes, scores, usage, costs, durations, and exception summaries. Detailed receipts, request IDs, exact event times, and host hardware metadata are kept locally. Earlier pushed commits still contain accounting metadata; this change does not rewrite Git history.
Validation: all 20 original ATIF files passed validation; minimization preserves every task's recorded results and exact token/cost sums. Bilingual commands, JSON examples, result tables, local links, archive checksum, and
git diff --checkwere verified. The changes contain no agent implementation changes.