You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Implements tool-output-consumption-rate — Tier 2 rank 12, the next unimplemented grader in .github/workflows/shared/graders/README.md (walking tier-first, then rank-within-tier: Tier 1 ranks 5-11 and Tier 2 ranks 1-4 were already done, making rank 12 the next one).
Why distinct from existing built-in graders: tool-output-consumption-rate measures whether tool call results were actually used downstream, not just whether calls succeeded. The built-in tool-success-rate grader only judges call success/failure; it says nothing about whether a successful result was ever referenced by a later action. This grader closes that gap by computing the fraction of IR observations[] entries with a non-empty consumedByActionIds, surfacing wasted exploration (tool calls whose output was fetched but never used).
Required IR fields: observations[] (specifically observations[].consumedByActionIds, as populated per the Trajectory IR build procedure step 4 in trajectory-ir.md).
Output contract: returns { value, unit: "ratio", details } with value: null / not-applicable message when the trace lacks an observations[] array.
See shared/graders/README.md for the full catalog — 8 of 25 graders now implemented.
Caution
Protected files were modified in this change.
This pull request is in
request_reviewmode and requires explicit human scrutiny before merge.Protected files:
README.md,.github/workflows/shared/graders/README.md,.github/workflows/shared/graders/tool-output-consumption-rate.mdImplements
tool-output-consumption-rate— Tier 2 rank 12, the next unimplemented grader in.github/workflows/shared/graders/README.md(walking tier-first, then rank-within-tier: Tier 1 ranks 5-11 and Tier 2 ranks 1-4 were already done, making rank 12 the next one).Why distinct from existing built-in graders:
tool-output-consumption-ratemeasures whether tool call results were actually used downstream, not just whether calls succeeded. The built-intool-success-rategrader only judges call success/failure; it says nothing about whether a successful result was ever referenced by a later action. This grader closes that gap by computing the fraction of IRobservations[]entries with a non-emptyconsumedByActionIds, surfacing wasted exploration (tool calls whose output was fetched but never used).Required IR fields:
observations[](specificallyobservations[].consumedByActionIds, as populated per the Trajectory IR build procedure step 4 intrajectory-ir.md).Output contract: returns
{ value, unit: "ratio", details }withvalue: null/ not-applicable message when the trace lacks anobservations[]array.See
shared/graders/README.mdfor the full catalog — 8 of 25 graders now implemented.Note
This was originally intended as a pull request, but the git push operation failed.
Original error: The process '/usr/bin/git' failed with exit code 1
Workflow Run: View run details and download bundle artifact
The bundle file is available in the
agentartifact in the workflow run linked above.Create the pull request manually