Causinator 9000 CI Diagnosis

Version updated for https://github.com/sylvainsf/causinator9000 to version v1.1.0.

  • This action is used across all versions by ? repositories.

Action Type

This is a Docker action.

Go to the GitHub Marketplace to find the latest changes.

Action Summary

The Causinator 9000 is a causal inference engine designed to identify and rank the most likely causes of infrastructure degradation by analyzing changes (mutations) and observed issues (signals) within a dependency graph. It automates root cause analysis by using Bayesian inference to compute causal probabilities, taking into account factors such as temporal proximity, dependency relationships, and resource-specific likelihoods. This enables teams to quickly pinpoint and address issues in complex, large-scale cloud environments.

What’s Changed

v1.1.0 — Smarter CI Failure Diagnosis

Major accuracy and usability improvements to the GitHub Actions source adapter and report generator. Validated against Prometheus (75% accuracy, 312 diagnoses) and Radius.

Highlights

  • Causal domains — Failures are now classified into independent domains: PR, Scheduled, Manual Dispatch, and Release Validation. Scheduled/LRT test failures are no longer buried under PR noise.
  • Commit ancestry — For scheduled and dispatch runs, C9K walks the commit history back to the last successful run, creating multiple competing causes with proper temporal decay. No more flat 88% confidence on everything.
  • Historical flaky rates — Per-workflow pass/fail ratios calibrate the flaky-test prior. A 40%-flaky workflow gets a different diagnosis than a 5%-flaky one.
  • Full log classification — --fast mode removed as default. Failure logs are downloaded and classified into specific signal types (LintFailure, AzureAuthFailure, UnitTestFailure, etc.) instead of generic TestFailure.
  • Rich reports — Per-domain sections, linked SHAs, numbered run links, signal types, and failed step names in all tables. Every failure shows what broke alongside what C9K thinks caused it.

New features

  • --exclude-workflow flag to filter noisy workflows (e.g. PR checklists)
  • WorkflowConfigFailure latent node for project-specific workflow issues
  • RunnerFailure latent node for GitHub Actions runner/provisioning failures
  • CI retrigger detection (empty commits, “retry”, etc.)
  • Fallback to GitHub jobs API when failure logs aren’t downloadable
  • POST /api/clear now accepts {"clear_graph": true} for full topology reset

Accuracy improvements

  • Azure auth patterns expanded (catches Login to Azure step failures)
  • Flaky-test timestamp fixed — no longer resets decay clock per failure
  • Commit class temporal decay half-life: 30min → 8 hours
  • Cross-domain dedup prevents PR and scheduled test failures from hiding each other

Infrastructure

  • Dockerfile: fixed config not available during Rust compile stage
  • Release workflow: container build + tags only (Marketplace publish via UI)

Container

docker pull ghcr.io/sylvainsf/causinator9000:1.1.0