test(episodes): characterize mixed admission boundaries

[stacknil]

Summary

• add a minimal three-candidate control whose adjacent core contrasts are [18x pass, 1x fail]
• require the aggregate admission to reject the unchanged v1 selection while retaining both local boundary decisions
• document that partial adoption needs a separate boundary-driven partition and identity contract

Why

Candidate-v2 exposes one segment-level admitted value plus pair-level contrast evidence. The missing mixed case left the aggregate meaning ambiguous: one passing boundary must not make an otherwise mixed selection admissible, while one failing boundary must not erase the passing local evidence.

How to validate

• python -m unittest discover -s tests -p 'test_episode*.py' — 44 passed
• python -m ruff check . — passed
• ctest --test-dir build -C Debug --output-on-failure — 7/7 passed
• git diff --check origin/main...HEAD — passed
• gitleaks git --log-opts="origin/main..HEAD" --redact --no-banner — no leaks
• mutation proof: changing aggregate all(...) to any(...) makes the new focused test fail with True is not false

Risk / rollout notes

• Design decision: admitted means the complete v1 selection may be adopted unchanged; it is not a per-candidate validity verdict.
• Main risk: a future consumer could misread aggregate false as permission to discard every selected candidate or the passing boundary.
• Compatibility impact: test and ADR only; no evaluator, fixture, schema, detector, CLI, or loglens.report.v3 behavior changes.
• Rollback path: revert the two characterization commits; no artifact or data migration is required.
• Follow-up: any partial acceptance must define boundary partitioning, merged evidence, and stable finding identity in a separate contract.

[stacknil]

Final diff review after CI: no findings.

• Scope remains test + ADR only; no evaluator, fixture, schema, detector, CLI, or report-v3 changes.
• The mixed control is explicitly one activity segment with three selected candidates.
• Exact local evidence remains [18x pass, 1x fail]; aggregate false means only that the full v1 selection cannot be adopted unchanged.
• The all -> any mutation fails the focused test, confirming the whole-selection gate is observable.
• Rollback is the three test/documentation commits with no migration.