Add test coverage as 4th Copeland scoring criterion

[that-github-user]

Summary

Copeland scoring currently has 3 criteria: tests passed, convergence, files changed. The "fewer files" criterion penalizes agents that add test files — Agent #5 beat Agent #1 in #80 because it changed 1 file (no tests) vs 2 files (code + 6 tests).

Proposed fix

Add a 4th criterion: test files added/modified. Count files matching *.test.* or *.spec.* patterns. More test files = better.

Also split criterion 3: count only non-test files for the "fewer files" comparison. This way:

1. Tests passed (pass > fail) — binary correctness
2. Convergence group size (larger > smaller) — consensus
3. Non-test files changed (fewer > more) — minimal scope
4. Test files added (more > fewer) — thoroughness

This ensures Copeland rewards both minimal code changes AND comprehensive testing.

Evidence

In PR #118, Copeland recommended Agent #5 (+4) over Agent #1 (0) solely because #5 changed fewer files. But #5 added zero tests while #1 added 6. Human override was needed.

Acceptance criteria

• Split filesChanged into testFiles and nonTestFiles in Copeland
• Add testFiles as 4th criterion (more = better)
• Update tests
• Re-run thinktank evaluate to measure impact

[that-github-user]

Fixed in #120. Copeland now has 4 criteria: tests passed, convergence, code scope (non-test files, fewer=better), test coverage (test files, more=better, capped at 3).

Anti-gaming safeguards:

• Test files only count when agent also changed production code
• Cap at 3 prevents score inflation from many test files
• 3/3 passing agents independently implemented the same anti-gaming logic (strong consensus)

Generated by thinktank Opus.