Skip to content
arula

Stage 630 minutesPairs301 · Tooled judgment

Run the plan

Run workbench run with tasks.json, apply only the approved repair scope and inspect the new implementation diff and run output.

You will produce

New implementation diff and run output

We will demonstrate

From tasks.json to a controlled new diff

Done when

The run follows the approved tasks and records every planned check as comparable evidence or an explicit omission.

Sibling repository../payments-validation-fixture836a75dThe preparation guide explains how to verify a newer revision before using it.

Orient

Why this stage exists

Run is the workbench execution step. It turns the plan into a changed implementation and recorded output.

A successful run is not approval. Every run creates another AI-authored diff that must return to Diagnose, Review and Eval.

Start from

What you start from

Work from these inputs only. Anything not on this list is either a later stage’s concern or a decision you do not own.

tasks.json
The approved plan from Stage 5.
Defect specs
The source defects that constrain the repair.
Allowed paths
The narrow files and tests the run may change.
Demonstrate

Execute tasks.json without inventing policy

One worked pass, so the shape of the work is visible before you do it on your own.

  1. Run workbench run with tasks.json and the allowed scope.
  2. Inspect the new patch and run output.
  3. Confirm logs contain refundId and attempt, not providerRequest.
  4. Confirm the timeout mutation is now caught and preserve idempotency as unresolved.

What this turns onGreen is an input to eval, not a release decision.

Practice

Do the work

  1. Execute tasks.json on a learner branch.
  2. Check that the repair did not add an invented refund-idempotency rule.
  3. Record any check that became silent or non-comparable.
  4. Hand the new diff and run output to Stage 7.
Output

You produce

Artifact

New diff and run output

  • Applied tasks
  • Patch
  • Test changes
  • Per-check state
  • Provenance
  • Unresolved omissions
Pressure-test

Challenge your own result

The run is green. Can the learner ship?

Show how this resolves

Not yet. The new diff still needs its own diagnosis, review and eval, including regression risk, newly silent evidence and the unresolved idempotency decision.

Gate

Ready to continue when

Readiness gate
  • The run follows the approved plan and scope.
  • Every planned check is comparable evidence or explicitly not run.

What transfersRun the plan as a controlled experiment, not as a green-status shortcut.