[claude]

Closes #146

What was missing

shaw_dieterich_distance_model, prune_distance_graph and probability_graph in source_modelling/rupture_propagation.py had no direct unit tests. The only coverage came through sample_rupture_propagation, which checks tree structure and never looks at edge probabilities. So dropping the metres-to-km / 1000 (the same kind of units bug as #124), loosening the strict < cutoff, or removing the 0.99 cap would all have passed the suite.

What changed

Tests only. I added nine tests to tests/test_rupture_propagation.py, following the issue's suggestions:

• shaw_dieterich_distance_model: returns 1 at distance == delta, returns exp(-1) at 4 km with d0=3, delta=1, and is clipped to 1 below delta.
• prune_distance_graph: removes edges beyond the cutoff, also removes an edge exactly at the cutoff (the comparison is strict), and keeps faults with no remaining neighbours as empty dicts.
• probability_graph: a distance of 4000 m gives a weight of exp(-1) (the model is evaluated in km), a distance of 0 m gives 0.99 rather than 1, and a node with no neighbours is still in the graph.

Verification

The library code is unchanged, so the new tests pass against it. To check they actually catch regressions, I made each of the three changes the issue names to the library code locally:

• dropping / 1000
• changing < to <=
• changing the 0.99 cap to 1.0

Each one made the matching test fail, and I reverted the library code afterwards. The full suite passes (810 tests).

🤖 Generated with Claude Code