Skip to content
Book a demoStart for free
Start for freeBook a demo

Code Review

Review Scorecard: code quality and blast radius

Every review now scores two things separately: the quality of the code, and how much of the system it can break. The Review Scorecard combines them into a merge confidence you can trace back to its cause.

Share on XRSS

Two axes, one deterministic matrix#

Code Quality runs from 1 to 5 and Blast Radius from Low to High. Merge Confidence comes from a fixed matrix of the two, so a small change with real problems can no longer read as safe to merge, and a clean change to a critical path is still flagged for a careful look. A basis line under the score names the findings that set it.

Update, October 2026: merge confidence now follows the findings alone, and blast radius sits beside it as context. The scorecard above comes from just before that change, so its merge confidence still reads code quality times blast radius. Merge confidence is re-scored when review threads are resolved.

Blast radius measured by dependency reach#

Blast radius counts how many files reference the symbols a pull request changes, with weight for auth, data, migration, infrastructure, API and billing surfaces. Test churn, common method names every class defines, and non-escalating public API changes are excluded, and the size is taken from the whole pull request rather than the last commit.

Reviews stay on the author's change#

When an author merges the base branch into their branch, the reviewer now reviews only the files the pull request owns, and findings outside the diff are no longer posted as top-level comments. Teams can also generate review guidelines from the comments their own reviewers leave, so the review learns the house style. See Code Review.

Improvements#

  • Code ReviewThe confidence score explains itself in the summary.
  • Code ReviewReview guidelines can be generated from human PR comments.
  • Code ReviewSearch the Active PRs list by pull request number, with column filters.
  • Code ReviewUp to 100 files per pull request get a full review.

Fixes#

  • Code ReviewDiagrams that fail to render are left out of the summary instead of posted broken.
  • Code ReviewFindings the reviewer produced are no longer dropped before posting.
  • Code ReviewA failed web search is no longer treated as evidence against a finding.
  • Code ReviewScores count only Entelligence findings, not comments from other review bots.
Code ReviewEllie CLI

The memory starts the day you connect a repo.

From that day on, every session, review and incident lands in the same index.

Start for freeBook a demo
  • No credit card required
  • No training on customer code
  • Deploy in your cloud, your own keys