Open Code Review: inspect review scope, filters and model work
Choose between OCR-managed review and delegation
Compare review approaches by scope preparation, model ownership and acceptance evidence instead of selecting solely from star counts or benchmark claims.
What you will learn
- Choose who owns the model call
- Keep deterministic checks in the workflow
- Write a conditional adoption decision
Before you start
- Git changes and merge-base comparisons
- Basic CLI use and model API credentials
Choose a review scope, explain exclusions, prepare a controlled pilot and distinguish source inspection from runtime evidence.
Key takeaways
- Managed and delegated review place model work in different systems.
- Existing tests remain part of acceptance.
- A selection decision should name its evidence and limits.
Choose who owns the model call
OCR-managed review is a fit when the team wants its review CLI to use an explicitly configured model provider. Delegation fits a workflow where an existing host agent should receive selected files and rules, then perform the reasoning with its own model access.
The operational difference includes credentials, accounting and the place where model failures are diagnosed. Choose one path for the first trial so that a missing provider setting or host permission does not get mistaken for a problem with Git scope selection.
Keep deterministic checks in the workflow
Compilers, tests and rule-based tools answer questions that a probabilistic reviewer cannot certify. Keep those checks when adding model review. Use the model’s comments as additional evidence that a human can confirm against the change and relevant tests.
For a repository with few known review failures, a small supervised pilot is easier to interpret than immediate organization-wide rollout. Select a task with observable acceptance criteria, such as finding a labeled bug without repeatedly flagging the neighboring harmless change.
Write a conditional adoption decision
Record the tested release, mode, provider and sample. State where the reviewer helped, where it missed defects, and whether the checking time was acceptable. A result from one language or change size should not silently become a claim about every repository.
Choose continued experimentation when important evidence is missing. For example, useful comments on public samples do not answer whether private code may be sent to the same endpoint. Licensing, data handling and account permission remain independent adoption conditions.
Decision guide
| Criterion | Option A | Option B |
|---|---|---|
| Best when | You need predictable behavior and easy auditing | You need adaptive optimization and have reliable telemetry |
| Main risk | May leave performance on the table | Can become difficult to explain or debug |
Implementation steps
- 1
Choose managed review or delegation based on model ownership and accounting.
- 2
Keep existing deterministic checks and define a small acceptance sample.
- 3
Write an adoption decision limited to the observed sample and authorized data.
Copy-ready example
{
"decisionRecord": true,
"mode": "choose managed or delegated",
"sample": "labeled synthetic changes",
"observedBenefit": null,
"missedDefects": null,
"dataApproval": false,
"adoption": "pending controlled trial"
}Frequently asked questions
Can OCR replace tests?
Its comments do not certify program behavior. Keep tests and other deterministic checks.
Which mode costs less?
That depends on provider or host usage, configuration and review workload. Measure the chosen workflow.
Sources
- Open Code Review / README.mdSource checked 2026-09-18
- Open Code Review / pages/src/content/docs/en/integrations/delegate.mdSource checked 2026-09-18
- Open Code Review / pages/src/content/docs/en/integrations/ci.mdSource checked 2026-09-18