NewClaude Code Trace
It evaluates agent behavior with a fixed set of interpretable metrics plus action-window questions, while minimizing or explicitly confirming session data before sending.
Added to Jevfast
NewHow it uses Jev
Claude Code Trace prepares a locally redacted session efficiency input and sends it to Jev after the user selects Analyse efficiency and confirms the payload. Ten base metrics use Noul questions and thinkingBalance uses a rubric Score; additional Noul questions are generated for bounded action windows. Typed probabilities and rubric positions are converted into dashboard metrics and trace-linked findings.
What Jev decides
Session efficiency checksExample answers · not a recorded Jev response · Source ↗
Question 1 · progressingEfficiently
YOUR APP
INSTRUCTION
Did the agent make steady, meaningful progress toward the user’s task?
STATE
Redacted EfficiencyInput: base questions assess session/task signals; generated window questions are limited to chunks of six actions, with up to twelve windows.
JEV · NOUL
YesNo
Question 2 · toolCallsUseful
YOUR APP
INSTRUCTION
Were the tool calls useful and proportionate to completing the task?
STATE
Redacted EfficiencyInput: base questions assess session/task signals; generated window questions are limited to chunks of six actions, with up to twelve windows.
JEV · NOUL
YesNo
Question 3 · likelyTaskCompleted
YOUR APP
INSTRUCTION
Does the trace indicate that the user’s requested task was completed successfully?
STATE
Redacted EfficiencyInput: base questions assess session/task signals; generated window questions are limited to chunks of six actions, with up to twelve windows.
JEV · NOUL
YesNo
App workflow
- Select a session
Choose Analyse efficiency for a Claude Code session.
- Review privacy payload
Inspect locally redacted data and explicitly confirm sending.
- Ask Jev
Send state and generated metric/window questions to System One.
- Inspect findings
Review dashboard metrics and trace annotations; re-analysis replaces prior result for the session.
Why it is interesting
It evaluates agent behavior with a fixed set of interpretable metrics plus action-window questions, while minimizing or explicitly confirming session data before sending.