Select a task
- TEST EXECUTION
- NOT EXPORTED
- TASK ID
- NOT EXPORTED
OPEN TEST / SINGLE SESSION
Inside the
repair loop.
A small repair bench for two agent workflows: solo implementation and implementation with a separate review. Open the brief. Read the patch. Inspect the test result.
Loading exported report…
BENCH READY
No run has been exported yet.
The viewer is ready. Results will appear here when the experiment produces an exported report. There are no sample scores standing in for measurements.
Inspect the experiment source01 / RUN RECORD
What was measured
| Workflow | Task checks | Workflow time | Runner time | Model | Tokens | Cost |
|---|
Runner time covers local compilation and tests. Workflow time can include scheduling and waiting, so it is not model speed. The workflows use different amounts of work: a review adds another pass. Passing public tests does not guarantee correctness.
02 / OPEN THE RECORD
Inspect the repairs.
Choose a workflow, then a task.
READING THIS LAB
A small experiment should make small claims.
Tasks and tests are visible to the agents. The runs belong to one session. Task outcomes, patches, and exported timings are evidence you can inspect; they are not a controlled study of model quality. Missing telemetry stays missing.
Read the method and reproduce the checks