[Srikar-Kolli]

Summary

Makes WeatherService.generate_synthetic_data() deterministic so tests and comparisons get the same weather every run.

• Uses a local np.random.RandomState(seed) instead of the global np.random. The default seed is 7, the same as the sweeps used, so existing sweep output is unchanged.
• Adds synthetic_weather_seed to LapsRaceSimulator and passes it through run_fixed_speed_sweep and run_strategy_sweep. The sweeps no longer call np.random.seed.
• Edge-case fixes:
  • A race starting exactly at 13:00 raised ZeroDivisionError.
  • Race windows shorter than the 15/30-minute smoothing kernels returned arrays longer than the time axis. A new _moving_average helper fixes this and gives the same values as before for normal windows.
  • An empty or reversed race window now raises a clear ValueError.
• Fixes two tests/planning/test_csv.py tests that were already failing on main. They used the old lap_distance_km field and fake args that no longer match main.py.

Testing

• pytest tests: 143 passed, 0 failed.
• New tests/test_weather.py (44 tests) covers:
  • Determinism: repeated calls, separate instances, and different seeds.
  • Global random state is neither read nor changed.
  • No network calls.
  • Seed 7 reproduces the old np.random.seed(7) output.
  • Array lengths and value ranges.
  • Edge cases: start at 13:00, start after solar noon, windows of 1 to 31 minutes, empty and reversed windows, fractional hours, and a full-day window.
  • Two end-to-end offline simulator runs that must give identical results.
• python3 -m compileall sim passes.

[Kamal-uva]

Great work on the deterministic weather implementation and thorough test coverage! The changes look good overall, but there's a merge conflict with the newer time-step functionality on main that we need to resolve before merging.

Changes needed:

1. sim/weather.py: Preserve resample_to_time_step() from main (including its existing implementation), since the PR branch was created before that functionality was added. Keep the calls to this helper in both fetch_weather_data() and generate_synthetic_data() so API and synthetic weather continue supporting configurable simulation time steps.

2. Preserve the PR's changes: Keep DEFAULT_SYNTHETIC_SEED, the local np.random.RandomState(seed), the _moving_average() helper, race-window validation, and seed propagation through LapsRaceSimulator and the planning sweeps. The final generate_synthetic_data() should generate deterministic weather, apply the new smoothing helper, and then return the arrays through resample_to_time_step().

3. tests/test_weather.py: Update the race() helper to include time_step_minutes=1.0, since the merged implementation now accesses that attribute. Add tests for non-default time steps (e.g., 5-minute aggregation and 0.5-minute interpolation), verifying that the returned time, GHI, and cloud-cover arrays have matching lengths and correct time spacing.

4. Regression testing: Confirm that the existing deterministic weather tests still pass, particularly the seed-7 legacy output comparison at the default 1-minute step. Also verify that the new time-step tests on main still pass and that no existing functionality is removed during conflict resolution.

After updating the branch with main and resolving the conflict, please run python3 -m pytest tests and python3 -m compileall sim to confirm everything works together.

The goal is to integrate the deterministic weather improvements without regressing the recently merged time-step functionality. Once that's resolved and tests pass, this should be ready to merge!