Week 6 studio: Review and dispose of findings
Bring a change from your first open-source contribution or your project. Keep the work in a dedicated directory following Tools and model access. This studio uses your change as the review subject, not as a request to run an unknown command supplied by a model.
Studio sequence
Format: individual first review, then a written exchange or a small discussion. Materials: your diff, its issue or specification, the relevant tests, a text editor, and the coding-agent seat or credit provided by the course. Use your agent directly. Select a small change whose checks you can already run. If access or execution fails, retain the failure and ask for support; a source trace can support an explicitly bounded claim, but it cannot be recorded as a model run or a passed runtime check.
1. Establish the review target (10 minutes)
Record the base revision, candidate revision, files in scope and the behavior that should change or stay the same. Write the four explanation answers before asking for review:
- What changed?
- Why was this change requested?
- Which input, caller or assumption could be wrong?
- What did you verify yourself?
If the request is ambiguous, write the two readings and the input that would separate them. Ask the person who owns the request for the decision; do not let a model fill the gap.
2. Review without AI (15 minutes)
Inspect the diff and the relevant source before reading any model output. Write at least one concrete review note, or document the cases you checked when no finding appears. Each note has a location, an input or path, an expected result, an observed or traced result, and a consequence. Choose a boundary case that the issue's example does not cover.
3. Request one model review (10 minutes)
Give the model only the contract, the diff and the files needed to understand the claim. Ask for locations, affected inputs and a check for each finding. Keep the result as text; do not apply generated code or execute a generated command without inspecting it. Record the agent, date, revisions and context files used; retain the comments needed to explain dispositions and reuse your existing AI-use note. No full chat export is required.
4. Dispose of every finding (25 minutes)
Make a table with one row per model finding:
| Finding | Disposition | Evidence | Next decision |
|---|---|---|---|
| Location and claim | Confirmed, refuted, or cannot tell | Input and result, source trace, or missing evidence | Accept, narrow, recheck, or stop |
For confirmed, reproduce the behavior or trace the exact code path. For refuted, record the counterexample or contract clause. For cannot tell, name what you inspected, what the files do not show, and the smallest safe check that would settle it. Add one row for a requirement the model did not mention; a silent review is not evidence that the requirement is satisfied.
5. Decide and hand off (10 minutes)
Compare your first review with the model review. State which evidence changed your view, what remains unknown, and whether the change is accepted, narrowed, sent back for another check, or stopped. The human with merge authority owns that acceptance decision. Attach the four explanation answers and the verification commands to the handoff; do not treat a model's confidence or a single green test as approval.
CP1
CP1 takes the first 30 minutes of the second session, on paper, without AI.
Participation and accessibility
All required material is plain text and can be completed with a keyboard. An approved written or asynchronous exchange can replace live discussion while preserving the same individual first review, finding dispositions and evidence. Keep one work record; do not copy private chat, personal data or credentials into it.