PR Risk Engine reference profile.
Will this pull request cause a production incident, and who should review it before it merges?
Overview
Reduce production incidents caused by risky changes.
Who owns this decision
VP Engineering, Head of Platform Engineering
What this Specialist Model actually does
Decision: Will this pull request cause a production incident, and who should review it before it merges?
The final taxonomy records 6 supporting or external components, but it doesn't promote an Oumi-owned specialist model for this engine.
Who uses this Specialist Model
- VP Engineering
- Head of Platform Engineering
- Head of Developer Productivity
- CTO
Where the specialist model sits
The final taxonomy records 6 supporting or external components, but it doesn't promote an Oumi-owned specialist model for this engine.
Feature pipeline joining diff, ownership, incident history, and test coverage
Supporting component
How Oumi connects to your stack →
Blast-radius and dependency-graph retrieval
Supporting component
Reviewer-routing model mapping risk and files
Supporting component
Reviewer-routing model mapping risk and files to the right human reviewers
Calibration layer
Supporting component
Calibration layer to keep predicted probabilities honest against observed rates
Code-specialized base model
Supporting component
Code-specialized base model for diff understanding
How Oumi connects to your stack →
Risk classifier fine-tuned on historical incident/rollback outcomes
Supporting component
How to evaluate and improve it
Evaluation
- Incident recall (share of true incident-causing PRs flagged high/critical)
- Calibration of predicted probabilities against observed incident rate
- Time-to-review for flagged vs. unflagged PRs
Feedback signals
- Incident opened after merge
- Rollback or revert
- Hotfix commit
- PagerDuty/on-call page tied to the change
- Customer-impact report linked to the change
- Reviewer overrides the risk score (accepts a "high" PR as-is, or escalates a "low" one)


