Investigate six flagged support-agent runs, follow the evidence through Opik, check the marked Python code, and submit a concise report for your group.
Use the workshop URL and group credentials provided by the facilitator to open your assigned Opik and Python code panes. Each group has its own environment, so this repository does not link to a specific group's instance.
A BA has flagged six suspicious support-agent runs before release. You are the engineer summarizing the debugging session for your manager. Most answers are inside Opik spans; four marked questions require checking the Python code.
Start with the case list on the workshop site. Open the matching trace, follow the named evidence, and use the code pane only when a question carries the Code label.
Open a new issue using the Mini workshop answers template. Keep one issue for your group. The issue body is setup only; submit answers as comments in that issue:
Case: 001
A: <short-answer-token>
B: <span-name>
C: <document-id>
D: <choice-token>
Your answers are plain GitHub issue text; you do not need to encrypt anything.
The automated checker grades each new non-bot comment once and reports the
cumulative score for your issue. New case answers are added to earlier answers;
a later non-empty retry replaces the earlier answer for that item. There are 24
answer items, so each correct answer adds 4.17 percentage points. Blank answers
are neutral; submitted wrong answers are marked check again.
Answers must be deterministic and short. Use the exact span name, document ID, metric name, or choice token requested by the question. Do not write prose explanations in the answer fields.
If you want to retry or add more answers, post another comment. Do not edit an old answer comment because edited comments are not graded again.