SCOUT

State-aware, cost-effective, online, uncertainty-corrected triage for flaky failures in distributed-database CI. This repository open-sources the distributed-DB CI simulator and the GitHub Actions metadata pipeline, together with their experiments.

Environment

uv sync

Pipeline Commands

make collect
make label
make features
make train
make eval

One-shot run:

make all

Validation:

make smoke
make replay
make test

GitHub Actions metadata-only dataset:

# requires GITHUB_TOKEN (read-only is enough)
PYTHONPATH=src uv run python -m state_triage.cli.main gha-collect --repo owner/repo
PYTHONPATH=src uv run python -m state_triage.cli.main gha-derive-sha --in-dir <runs_dir>
PYTHONPATH=src uv run python -m state_triage.cli.main gha-merge --in-dir <dir_a> --in-dir <dir_b>

CLI

PYTHONPATH=src uv run python -m state_triage.cli.main --help