Formalize collaboration workflow for human contributors

[GiggleLiu]

Summary

docs/paper/collaborate.md drafts a 5-step collaboration workflow. This issue tracks refinements discussed during review.

Current Draft

1. File an issue
2. Validate issue (Rule/Model criteria, source whitelist, check record)
3. Implement (includes documentation)
4. Review implementation (code quality, faithful to issue)
5. Agentic test (scenario-based user personas)

Clarifications Resolved

• Source whitelist = references listed in README (Garey & Johnson, Complexity Zoo, Compendium of NP Optimization Problems, ProblemReductions.jl, etc.), to be expanded over time.
• Documentation is part of Step 3 (Implement), not a separate phase. Step 4 review focuses on code correctness, not prose quality.
• Issue check records are saved in docs/reviews/history.md.

Proposed Improvements

1. Typo: "senarios" → "scenarios" (line 16)
2. Step 2 wording: "algorithm is useful, correct, detailed, and optimal" — consider rewording to "reduction is useful, correct, detailed, and optimal" for Rule issues, since the validated artifact is the reduction, not the algorithm.
3. Step 3 — Implement: currently one line. Consider expanding to mention:
   • Test coverage requirement (>95%)
   • Paper entries (problem-def / reduction-rule in reductions.typ)
4. Step 2 — Source whitelist: make the README reference explicit, e.g., "References must come from the whitelist in README.md § Acknowledgments / Related Projects, or equivalent peer-reviewed sources."
5. Step 2 — Check record: mention docs/reviews/history.md as the storage location.
6. Step 5 — Agentic test: the three personas are good. Could benefit from noting whether these are manual LLM sessions or automated test harnesses (to be decided).
7. Missing: no mention of CI gates (make check must pass before merge).

Action Items

• Fix typo
• Refine wording per above
• Create docs/reviews/ directory and seed history.md
• Decide on agentic test execution format (manual vs automated)