Signal sources

A signal source is an input an AI investigation reads when it runs. Each source produces a different kind of finding. Sources are additive: every investigation starts with your own incident history, and each integration you connect gives later runs more to work with. A signal source is read by an investigation — it is the opposite direction from an inbound alert integration, which pushes alerts into scaling.cloud. Connecting Sentry as a signal source, for example, lets investigations read your Sentry issues; it does not turn Sentry alerts into incidents.

​

Finding kinds

​

Past incidents

The built-in source, available to every entitled organization with no integration to set up. scaling.cloud matches the open incident against your resolved incidents by symptom and surfaces the closest matches as similar_incident findings, with a link to each prior incident and how it was resolved. This source only ever reads your own organization’s incidents.

​

GitHub — code changes

“What changed recently?” is the most common real root cause. Connecting the GitHub App lets investigations correlate your deploys, merges, and commits to the incident timeline and surface the likely culprit as a suspect_change finding. Changes are matched to an incident by time and component — a change to a repository that maps to one of the incident’s affected components, landing shortly before the incident opened, is a suspect. Monorepos map to different components by path, so a change is attributed to the component it actually touched.

​

Sentry — errors and releases

Sentry has already done the hard part of telemetry — grouping errors, capturing stack traces, and linking errors to the release and commit that introduced them. Connecting Sentry lets an investigation read that work and produce two finding kinds:

• root_cause — derived from a grouped error’s stack trace.
• commit_correlation — from a Sentry release’s linked commits, corroborating a GitHub suspect_change at commit precision.

​

Graceful degradation

Signal sources are independent. If a source is not connected, or a connected source can’t be reached during a run, the investigation simply skips it and reports what the other sources found — a missing source never blocks an investigation or holds up paging.