Contributing

The project is small on purpose: one problem, one cost function, one policy, one simulator, one reference implementation. Please keep it that way.

Ground rules

• No new features without a baseline. Anything that changes scheduling behaviour must be compared against reactive_fastest and oracle in examples/compare_policies.py and the numbers pasted in the PR.
• Numbers must be reproducible. Fix seeds, state the config, commit the JSON output under results/ only for tagged releases.
• Prefer deleting code. The cost function in policy.py is the paper. Complexity there needs a written justification in docs/algorithm.md.
• Format with ruff, test with pytest. CI runs both.

Where things live

Path	What
src/hazardserve/hazard.py	Survival estimation (Kaplan-Meier + conditional survival)
src/hazardserve/cost.py	Time and migration cost models
src/hazardserve/policy.py	The algorithm + baselines
src/hazardserve/simulator.py	Churn-driven discrete-time simulator
examples/	Runnable experiments
docs/	GitHub Pages (MkDocs Material)
paper/	Paper source and figures
PLAN.md	Phases, milestones, pivot thresholds

Good first issues

See PLAN.md Phase 1 and Phase 2 checklists; anything unchecked is open.