NEWS

mapfit 1.0.1 (2026-09-28)

Fixed the handling of left-truncated grouped data. breaks not starting at 0 add an interval whose count is unknown, and it was marked with NA after the conversion of missing counts to the internal sentinel. The estimation step then treated the interval as if it had been observed to contain no samples.
Missing counts (NA) and left-truncated data are now rejected in mapfit.group(). The estimation algorithm for MAP uses the number of arrivals in an interval as an array index, so an unknown count read past the end of the array instead of being marginalised out. Missing counts remain supported in phfit.group(), which is what the documentation has always described.
Fixed the exit rate of a CF1 after estimation. phfit.point() and friends updated the initial probabilities and the rates but left the exit rate at its initial value, so dphase() on an estimated CF1 returned a multiple of the density; the values did not integrate to 1. The degrees of freedom reported for a CF1 are recomputed from the estimated parameters as well.
cf1() no longer sorts the vectors passed as alpha and rate in the caller's environment.
Passing data to a model that cannot handle it now raises an error. In particular mapfit.point() with a gmmpp model, which is an algorithm for grouped data, silently returned a result object with no log-likelihood.