Test coverage gaps and weak assertions in privacy and integration paths

[cursor]

Summary

Unit tests cover privacy primitives and basic simulation smoke tests well (~43 tests), but several important paths lack coverage or use weak assertions that can pass without validating intended behavior.

Gaps

• No CLI / app.py integration tests
• No test asserting full token → threshold → dilution → broadcast → response → detection pipeline
• Attack CLI flags (e.g. --enable-deanon) not validated in simulation context
• plot_metrics() and _classify_detection() largely untested
• NetworkAggregator.evaluate_and_broadcast() full path untested in simulation

Weak assertions

• test_multiple_queries_still_noisy asserts error is not None but does not bound error (comment suggests ~200m intent)
• Spatial dilution tests use if sparse_cell is not None and may skip assertions silently

Impact

• Integration bugs (e.g. zone ID mismatch) can slip through
• Regressions in protocol behavior may go undetected

Suggested fix

• Add integration test for end-to-end privacy protocol
• Strengthen privacy test bounds where statistically stable
• Remove conditional skips or use fixtures that guarantee sparse/dense cells

Acceptance criteria

• Integration test covers anomaly clustering → broadcast → agent response
• Weak assertions replaced with meaningful bounds or marked as flaky with rationale
• Optional coverage reporting in CI