M4.1a — README and contributor guide

[Ferrari-72]

Parent

Spawned from #8 (M4.1 tracking issue).

Objective

Write a README that explains:

• What this benchmark measures and why (counterexample certificate format, three violation types)
• How to add a model (what runner to implement, what results file to produce)
• How to run a session locally (make demo, env vars needed)
• How to read the metrics (bugs/Ktok vs bugs/$, why bugs/Ktok is the primary ranking)

Definition of done

• README.md exists at repo root with all four sections above
• A CONTRIBUTING.md documents the certificate JSON schema and the verification workflow
• make test-unit still passes

Dependencies

Depends on #1–#7 (all prior milestones).

[Ferrari-72]

Completed — commit c8f97b6

What was done

README.md (new, repo root) — four sections:

1. What this measures — round-trip failure definition, 3 violation types table
2. How to add a model — AgentRunner interface + results file format
3. How to run locally — env vars, make demo, all key make targets table
4. How to read the metrics — bugs/Ktok (primary, model-agnostic) vs bugs/$

CONTRIBUTING.md (new, repo root):

• Full certificate JSON schema with required fields per violation type
• 3 violation type definitions
• Verification workflow: make verify-calibration, make verify-judgment, make test-unit
• 3 canonical fixtures and their expected verdicts
• Step-by-step guide for adding a new model and running CI locally

test_docs.py — 8 tests, @pytest.mark.judgment

Test	Verifies
test_readme_exists	README.md present at repo root
test_readme_has_what_section	contains "violation" or "counterexample"
test_readme_has_how_to_add_model	contains "AgentRunner"
test_readme_has_run_locally	contains "make demo" or "REPO_DIR"
test_readme_has_metrics_section	contains "bugs/Ktok"
test_contributing_exists	CONTRIBUTING.md present
test_contributing_has_certificate_schema	contains "source", "bundle", "violation"
test_contributing_has_verify_calibration	contains "verify-calibration"

Test results

141 passed, 2 skipped   (make test-unit)