Decay check

Does this repo still install and run today? The agent-free reproducibility signal — the same naive_runs check that found 85% of a random sample of recent Bioinformatics tools don't run on their own. Available as a CLI and a GitHub Action. No agent, no API key.

CLI

pip install lazarus-bio
lazarus decay-check https://github.com/owner/repo

It shallow-clones the repo, installs from its own files (conda / pip / R — first manifest wins, no repair), runs a shipped example (or the README's usage block, or a bare import), and prints a verdict with a reason code:

https://github.com/owner/repo
  DECAYED — does not run on its own today
  stage/reason: install/pip_install_failed  (python, host, 42s)
  → Lazarus can revive it: https://doctordean.github.io/lazarus/

Options: --sandbox host|docker · --fail-on-decay (non-zero exit for CI) · --json · --timeout <s> · --docker-host ssh://… (docker sandbox).

Two sandboxes:

• host (default) — runs on the current machine; Python installs into a throwaway venv, R into a temp library. Light, no images. A fresh CI runner is the clean environment.
• docker — runs in a fresh continuumio/miniconda3 / rocker/r-ver container. Strict parity with the published benchmark. Verdicts are less sensitive to the host toolchain.

GitHub Action

Reproducibility canary — watch your own repo

# .github/workflows/decay-check.yml
name: decay-check
on:
  schedule: [{ cron: "0 6 1 * *" }]   # monthly
  workflow_dispatch:
jobs:
  decay:
    runs-on: ubuntu-latest
    steps:
      - uses: DoctorDean/lazarus/actions/decay-check@v0.3.0
        with:
          fail-on-decay: true          # a red build = "your code no longer runs"

Check any repo

      - uses: DoctorDean/lazarus/actions/decay-check@v0.3.0
        with:
          repo: https://github.com/owner/repo
          fail-on-decay: false         # informational; never fails the job

Use the outputs

      - id: decay
        uses: DoctorDean/lazarus/actions/decay-check@v0.3.0
      - run: echo "runs=${{ steps.decay.outputs.naive-runs }} reason=${{ steps.decay.outputs.reason }}"

Inputs: repo (default: this repo) · sandbox (host|docker) · fail-on-decay · timeout · python-version (default 3.11) · version (pip spec). Outputs: naive-runs · stage · reason. Every run writes a job summary; a decayed result links back here.

Reason codes

naive_runs	stage / reason
✅ true	example/ran_ok, example/imports_only
❌ false	install/{pip_requirements_failed, pip_install_failed, conda_env_failed, R_install_failed, no_install_manifest}, example/{example_exit_N, no_example_found}, install/{needs_conda, needs_r, venv_failed}, timeout/hard_cap

What it does and doesn't mean

A pass means "installs and a shipped example runs to exit 0" — not that the science is correct. A fail in host mode reflects the runner's toolchain (pin python-version, or use sandbox: docker for host-independent verdicts). It only tries the repo's own artefacts — no repair — which is exactly what makes it a fair measure of decay, and the counterfactual Lazarus is measured against.