Allow adjusted calibration when censoring is absent

[uriahf]

What changed

• allows censoring_heuristic="adjusted" in create_calibration_curve_times() when the supplied outcome data contain no censored observations
• continues to reject censoring_heuristic="adjusted" when any censored observations are present
• applies the same rule in the internal time-dependent calibration loop so allowed no-censoring inputs are not silently skipped
• leaves competing_heuristic="adjusted_as_censored" behavior unchanged
• adds regression tests for both public entry points covering:
  • adjusted censoring with no censored observations: allowed
  • adjusted censoring with censored observations: clear ValueError
  • competing-as-censored: still rejected

Why

When there are no censored observations, the censoring heuristic is statistically irrelevant. Rejecting "adjusted" purely because of its label creates an unnecessary API asymmetry with the other _times curves. Actual adjusted calibration under censoring remains unsupported and is deliberately not implemented in this PR.

Validation

The focused tests/test_calibration_times.py suite passed after applying both the public validation change and the matching internal-loop change.

[github-actions]

PR Preview Action v1.8.1
Preview removed because the pull request was closed.
2026-08-12 11:49 UTC