Explanation
Findings & severity
What each reasoning plane emits and how severities aggregate into a verdict.
A finding is the atomic output of a reasoning plane — a single observation about the PR with a severity, a location, and an explanation.
Anatomy of a finding
type Finding = {
id: string; // ulid
plane: Plane; // which plane emitted it
severity: "info" | "warn" | "error"; // gradient, not a binary
location: { file: string; lineRange: [number, number] };
title: string;
rationale: string; // why the plane raised it
related?: string[]; // ids of related findings
};Severities
| Severity | Surfaces as | Blocks merge? |
|---|---|---|
info | A note on the PR comment | No |
warn | A flag on the PR + email | Configurable |
error | A blocking finding | By default, yes |
How findings roll up
The Orchestration stage aggregates findings into a single tier using the highest severity present, weighted by plane confidence. Tenant policies can override the rollup — for example, "treat any Security warn as error."
Related
Was this page helpful?