[hmgaudecker]

What changes

• create_model (baseline and ACA) and create_benchmark_model take final_age, the first age at which everybody is dead. It defaults to MODEL_CONFIG.end_age (96), so existing callers build the same model as before.
• config.model_ages(final_age=...) returns the ages from 51 through final_age. It rejects a final age that leaves no living period or goes past 96, because the age-indexed data stop at 96.
• Nothing that depends on the ages is fixed at import time any more. The age stages, the default entry initial nodes, the next-age lookup inside the regime transitions, the transition ages, and the edges toward dead are all computed from the ages passed in. INITIAL_NODES is replaced by build_entry_initial_nodes(ages=...). initial_nodes=None (the new default) admits the entry regimes at ages 51–60 that come before the final age.
• Age-indexed parameters are not trimmed in aca-model. pylcm drops Series rows for ages that are not on the grid. Arrays indexed by period are read only for the periods the model has, and a shorter horizon uses a prefix of them.
• Regimes whose stage starts at or after the final age cannot be reached. Their schedule is a ByAge that holds only the dead cell as a default, and no period reads it.
• tests/test_medical_cost_timing.py builds ages 51–53 instead of the full horizon. Its assertions are unchanged.
• New tests:
  • a model with final_age=53 has n_periods == 3;
  • model_ages(final_age=53) == (51, 52, 53), and out-of-range final ages are rejected;
  • a survivor of age 52 moves to dead when the model ends at 53;
  • entry regimes have transition ages (51, 52).
• tests/test_nbegm_model_creation.py is marked long_running.
• CI installs pytest-xdist and runs pytest -n 4 --durations=20 tests. The -m "not long_running" addopts setting is unchanged.

https://claude.ai/code/session_01Hs8mfXPYyEa24KaAp6BwCW