test(cognitive-load): pin cache eviction behavior for both in-memory caches

[tongshu2023]

Summary

Addresses the remaining acceptance item on #30.

The two caches named in the issue already enforce hard bounds in the current code (the issue predates the fix): cognitive_load_nlp._analysis_cache caps at 500 and flushes the oldest half when full (under an asyncio.Lock), and cognitive_load_calibrator._cache caps at 200 and evicts the least-active (lowest message_count) baseline. So of the acceptance criteria, the first two are already satisfied — what was missing was the unit test verifying eviction behavior, without which a regression back to unbounded growth would land silently.

Changes

tests/test_cognitive_load_cache_bounds.py (6 tests):

NLP analysis cache

• never exceeds its cap under sustained distinct messages
• crossing the cap drops the oldest entries and keeps the newest
• a cache hit returns the cached object without re-analysis

Calibrator baseline cache

• never exceeds its cap
• eviction removes exactly the least-active baseline and admits the newcomer
• same user always receives the same baseline instance

The fixture monkeypatches the caps small (the eviction logic reads the module attribute at call time), keeping the suite at ~6s instead of ~3min with production-size caps.

Tests

test_cognitive_load_cache_bounds.py + test_cognitive_load_calibrator.py: 23 passed locally.

Acceptance criteria

• Both caches have a defined max size — already in code (500 / 200)
• Eviction happens automatically when max size is reached — already in code
• Existing tests pass
• Add a unit test verifying cache eviction behavior — this PR

🤖 Generated with Claude Code