Summary

Replace coverage-driven tests with test of behavoir and correctness.

• Replace low-value tests (shape-only assertions, private method tests) with tests that validate real-world behavior
• Consolidate shared fixtures in conftest.py, remove 5 dead fixtures, add 4 reusable fixtures (gen_int_rv, hierarchical_normal_noise, hierarchical_normal_noise_tight, hierarchical_infections)
• Extract shared non-fixture test code (ConcreteMeasurements, create_mock_infections) to new test/test_helpers.py module
• Test base class no-ops and validation stubs in file test/test_interface_coverage.py

Changes by file

test/conftest.py

Removed 5 dead fixtures, added 4 shared fixtures.

Dead fixtures removed (never used by any test file):

• medium_delay_pmf
• counts_process_medium_delay
• counts_process_realistic
• simple_shedding_pmf
• medium_shedding_pmf

Dead fixtures removed (replaced by new shared fixtures):

• sensor_mode_rv, sensor_sd_rv
• sensor_mode_rv_tight, sensor_sd_rv_tight

New shared fixtures added:

• gen_int_rv -- COVID-like 7-day generation interval
• hierarchical_normal_noise -- standard HierarchicalNormalNoise with VectorizedRV
• hierarchical_normal_noise_tight -- tight variance version for near-deterministic testing
• hierarchical_infections -- pre-configured HierarchicalInfections instance

test/test_helpers.py (new file)

Shared non-fixture test code extracted here (since test/ is a package, conftest.py cannot be directly imported by test modules):

• ConcreteMeasurements class (concrete implementation of abstract Measurements for testing)
• create_mock_infections() function (generates spike, constant, or decay infection curves)

test/test_temporal_processes.py

Added:

• test_ar1_mean_reversion -- verifies AR1 trajectories revert toward zero
• test_differenced_ar1_trend_persistence -- verifies DifferencedAR1 produces persistent trends

test/test_hierarchical_priors.py

Strengthened:

• Shape-only tests now include value assertions (centering near zero for Normal/StudentT, positivity for Gamma)

test/test_hierarchical_infections.py

Added:

• test_infections_are_positive -- verifies all infection values are positive (epidemiological invariant)
• test_shape_and_positivity_across_subpop_counts -- parametrized over K=1,3,6 with positivity and weighted-sum checks

Now uses conftest fixtures (gen_int_rv, hierarchical_infections) instead of local fixtures.

test/test_observation_measurements.py

Added:

• test_log_scale_correctness -- verifies output equals log(convolved infections) with known inputs
• test_sensor_bias_differences -- verifies hierarchical noise produces sensor-specific biases

Strengthened:

• test_sample_shape now also verifies sensor variation and log-scale output

test/test_observation_counts.py

Added:

• test_convolution_hand_computable -- 1000 infections on day 10 with PMF [0.5, 0.5] should produce 500 on days 10 and 11
• test_observation_passthrough -- providing obs=known_values returns those exact values
• test_poisson_mean_approximation -- mean of 50 Poisson samples approximates expected rate

Strengthened:

• test_zero_infections now asserts predicted == 0 (not just shape)
• test_small_infections now asserts plausible range on predicted values
• test_sample_returns_correct_shape now includes non-negative value assertions

test/test_pyrenew_builder.py

Added:

• test_prior_predictive_multi_signal -- uses numpyro.infer.Predictive to draw prior predictive samples from a builder-constructed model; verifies shapes and positivity

Strengthened:

• test_run_with_population_structure now also asserts all infection samples are positive

test/test_interface_coverage.py (new file)

Tests interface contracts (repr, validate(), infection_resolution(), get_required_lookback()) across implementing classes via parametrized tests