Skip to main content
AI Development

Baseline Copilot code review address rate before Adaptive

GitHub's Repository insights page shows how often teams act on Copilot code review comments. Record those address rates before you switch on Adaptive.

Sean McLellan profile photo

Sean McLellan

Lead Architect & Founder

7 min read
Editorial still life on a light wooden workbench of three brass magnifying loupes of increasing size, a copper selector lever on a wooden base pointing at the middle loupe, and a stack of cream paper behind them with two copper check-mark tabs
Constructed diagramIllustration: three review depths and a selector, for Copilot code review choosing Lite, Balanced, or Max. Not product UI.

GitHub added a Repository insights page for Copilot code review on November 17, 2026, along with a public preview of Adaptive review effort. Repository insights shows how often developers act on Copilot's review comments in each repository, broken down by severity and by review effort. Adaptive lets Copilot choose Lite, Balanced, or Max for each pull request. Before you make Adaptive the default, record a few weeks of address rates so you have something to compare against.

The order matters because the two features interact. Adaptive changes which effort level each review uses, and Repository insights files every Adaptive review under the level it picked. If you switch first, your Lite, Balanced, and Max numbers will mix your old configuration with Copilot's choices, and you lose the cleanest comparison you will get.

What is the Copilot code review address rate?

The address rate is the share of evaluated Copilot code review comments that developers acted on. GitHub's metrics reference defines it as addressed comments divided by evaluated comments, with a separate rate for high-severity comments.

GitHub decides what counts. When a pull request merges, an AI model checks each published Copilot comment to judge whether the issue it raised was addressed. A comment counts as addressed when a developer applies the suggestion, asks Copilot to fix it, pushes a fix (by hand or with any agent), or accepts the feedback and explicitly defers the fix to a later pull request.

The page shows four summary signals for each repository:

  • Overall address rate: addressed comments divided by evaluated comments.
  • High-severity address rate: the same calculation for high-severity comments only.
  • No human follow-up request: merged pull requests that Copilot reviewed where no human reviewer requested changes after Copilot's first review.
  • Merged without Copilot review: merged pull requests that Copilot did not review.

The page also breaks the address rate down by severity, review effort, and customization (custom instructions, agent skills, MCP servers, or none), and ranks addressed comments by category, such as Bug, Security, or Performance.

Who can see Repository insights?

Anyone with read access to an organization-owned repository can open the page from the repository's Insights tab. The Copilot code review section appears only when the organization has turned on both the Copilot usage metrics policy and the Copilot code review policy, according to GitHub's setup guide.

The same numbers are available in the enterprise and organization repos-1-day reports of the Copilot usage metrics API. API access is limited to enterprise owners, billing managers, organization owners, and custom roles with the View Copilot Metrics permission. The repos-1-day report is the same one that carries GitHub's pull request review-stage timings. Those timings cover how long human reviews take. Repository insights covers what happens to Copilot's comments.

How Adaptive chooses a review effort

Adaptive looks at each pull request's complexity and risk, then reviews it with Lite, Balanced, or Max. The review shows which level Adaptive picked. Two limits keep it bounded:

  • Adaptive never picks a level above the maximum review effort set for the enterprise, organization, or repository.
  • Balanced stays the default until someone chooses Adaptive. GitHub's configuration docs say the effort level "defaults to Balanced, not Adaptive."

Adaptive is in public preview for Copilot Pro, Pro+, Max, Business, and Enterprise. The Max review effort level is also in public preview.

Where you set Adaptive matters. GitHub's code review concepts page lists the order Copilot checks, and the first setting that applies wins:

  1. An effort chosen when the review is requested
  2. An effort already used on this pull request
  3. The requestor's personal Copilot review effort (for a new pull request, the author)
  4. The repository setting
  5. The organization setting
  6. The enterprise setting
  7. GitHub's default, Balanced

A developer's personal setting outranks the repository and organization defaults. Setting Adaptive at the organization level doesn't move people who already picked a level for themselves, and one developer who turns on Adaptive personally changes their own reviews even if the repository says Balanced. The maximum review effort setting caps Adaptive no matter where Adaptive is turned on.

Diagram titled How Copilot code review picks a review effort. Left panel, First setting that applies wins, lists seven steps in order: effort chosen when the review is requested, effort already used on this pull request, requestor's personal setting (highlighted), repository setting, organization setting, enterprise setting, GitHub default Balanced. Right panel, If the setting is Adaptive: Copilot picks Lite, Balanced, or Max by pull request complexity and risk; never above the maximum review effort set for the enterprise, organization, or repository; Repository insights counts the review under the level Adaptive picked.
Constructed diagramConstructed diagram based on GitHub's code review documentation and November 17 changelog (November 2026). Not product UI.

What Adaptive can do to your AI credit spend

Effort level drives cost. GitHub estimates an average review at about $0.25 in AI credits with Lite, $1.50 with Balanced, and $10 with Max. The typical ranges are $0.05 to $1, $0.25 to $5, and $1.50 to $25. GitHub calls these estimates only, and they exclude GitHub Actions minutes.

On a repository that runs Balanced today, Adaptive can lower spend if it moves routine pull requests to Lite. It can raise spend if it sends a meaningful share to Max. The docs don't say how often Adaptive picks each level, so measure it on your own pull requests. If Max is too expensive for a repository, set the maximum review effort to Balanced, and Adaptive will choose only between Lite and Balanced there.

How to set a baseline before you switch

BaristaLabs recommends a short, bounded pilot rather than an organization-wide switch:

  1. Turn on the data. Confirm the Copilot usage metrics and code review policies are on. Data can take a few days to appear, and "Waiting on your data" does not mean the repository had no activity.
  2. Record the current numbers. For two or three busy repositories, note the overall and high-severity address rates, the rate for each effort level, and the merged-without-review share for at least two weekly periods. The monthly trend on GitHub.com goes back to August 2026, which helps if you have that history.
  3. Pilot Adaptive on one or two repositories. Set it as the repository default, set the maximum review effort you are willing to pay for, and ask pilot developers to clear any personal effort setting so the repository default applies.
  4. Compare like with like. After the same number of weeks, check the high-severity address rate first, then the effort mix and AI credit usage. If you pull the API data, add up addressed and evaluated counts across days before dividing. Do not average daily rates.
  5. Read examples before you decide. Open a sample of unaddressed high-severity comments from both periods. A drop in the rate is a reason to read comments, not proof that the reviews got worse.

GitHub's metrics reference is explicit: "Differences between effort levels or customization groups are descriptive. They do not prove that increasing effort or adding customization causes a higher address rate or better code quality." A repository with no evaluated comments has no rate at all, which is different from a measured 0%. In our view, small repositories may never produce enough high-severity comments for a fair comparison, so treat them as coverage checks rather than pilots.

Use the coverage gap too

Check the merged-without-Copilot-review signal alongside the address rate. It shows how many merged pull requests Copilot did not review. GitHub notes that an "Automatic review is off" message describes configuration, not the number of pull requests that missed a review, so check the actual uncovered pull requests before you add a ruleset.

If you widen coverage, remember who pays. AI credits for automatic reviews go to the pull request author by default, as covered in our post on separating who pays from who can request reviews. Adding Adaptive and wider coverage at the same time makes it hard to tell which change moved the bill. Change one, measure, then change the other.

What this means for engineering leads

For a team already paying for Copilot code review, Repository insights puts numbers on whether those reviews help: which comments get acted on, how many merged pull requests skip Copilot review, and how outcomes differ by effort level. Adaptive could cut review cost or raise review depth, but GitHub's documentation does not say how it distributes pull requests across levels. A baseline taken before the switch is the clearest way to tell which one happened in your repositories. If you set your default during the move to Balanced in September, check whether those settings still match what you want before adding Adaptive on top. Teams that want help scoping the pilot can find our approach on the AI consulting page.

Sources and scope

These sources describe how the metrics and settings work. They do not show how Adaptive distributes real pull requests across effort levels, and BaristaLabs has not measured that. Cost figures are GitHub's estimates and will vary with pull request size and custom instructions.

Copilot code review pilot

Measure Copilot code review before you change the default

Share how your repositories configure Copilot code review today. BaristaLabs can help pick pilot repositories, set effort caps and personal-setting rules, and define the address-rate and AI credit numbers to compare before and after Adaptive.

Your current code review settings and a list of your busiest repositories are enough to start.

Turn this idea into a pilot

Which workflow should go first?

Use the readiness check to compare impact, effort, risk, owner, and next step before requesting a review.

  • 3-5 minutes
  • Deterministic score
  • No sensitive data
Check workflow readiness

Practical AI Workflow Notes

Want more practical AI operations ideas?

Get short notes on applying AI inside real small-business workflows — from document handling and customer follow-up to internal reporting, compliance, and automation guardrails.

A useful next step if you’re still exploring and not ready to request a 20-minute workflow assessment.

Occasional emails. Practical workflow guidance only. Unsubscribe anytime.