v0.8.2

0.8.1

Compatibility patch; no library-code changes.

• Cap the actuarialpy requirement at the next minor: >=0.46,<0.47.
• Declare Python 3.14 support in the classifiers.
• Fix one test fixture that used freq="ME" (pandas>=2.2 only) to literal
  month-end dates, so the suite runs at the declared pandas floor. Found by
  the new minimum-dependency CI job.

v0.8.1

0.8.1

Compatibility patch; no library-code changes.

• Cap the actuarialpy requirement at the next minor: >=0.46,<0.47.
• Declare Python 3.14 support in the classifiers.
• Fix one test fixture that used freq="ME" (pandas>=2.2 only) to literal
  month-end dates, so the suite runs at the declared pandas floor. Found by
  the new minimum-dependency CI job.

v0.8.0

0.8.0

project() accepts an ExperienceSet (routes to tab). Test fixtures
updated for the Measures -> Source rename in actuarialpy 0.46.

v0.7.0

[0.7.0] - 2026-07-11

Project from the canonical Experience (breaking).

• Remove ClaimExperience; the projection pipeline consumes
  actuarialpy.Experience directly.
• Add project(exp, ...): the single projection entrypoint. Named parameters
  are the pipeline phases (completion, seasonality, trend, credibility,
  complement); record grain defaults to the Experience's bound dimensions,
  and the claim-type dimension is inferred as the grain column absent from
  the exposure frame (pass claim_type= when ambiguous).
• Add base_rates(exp, ...) and prepare_experience(exp, ...) as public
  steps of the same pipeline; ClaimProjection is unchanged as the engine
  and ClaimProjection.from_experience is removed in favor of project.
• Requires actuarialpy>=0.45.
• project() coerces raw assumption values (scalar trend and credibility,
  Series/mapping completion and seasonality) into assumption objects, and
  resolves the exposure frame's column from the bound exposure role name.
• project() consumes wide from_tables Experiences directly: the recorded
  pivot melts itself into the claim-type dimension, non-pivot expense columns
  are announced as excluded, and ambiguity asks for claim_type=.
• seasonality="estimate" fits factors from the bound history via
  estimate_seasonality; completion="estimate" explains what development
  columns estimation needs instead of guessing.

v0.6.5

[0.6.5] - 2026-07-11

Changed

• ExpenseProjection's docstring now states three behaviours that were
  always true but only discoverable by reading the validation code: trend
  may be keyed by expense type (a zero-trend type is a contractually flat
  fee); the claims/premium inputs are keyed value streams, so a
  percentage of a claims subset or of another expense run's output is the
  same mechanism fed a different table; and percentage bases are not
  re-prorated by active_fraction.
• examples/expenses.py now exercises the keyed expense trend: a
  contractually flat per-exposure network fee projects at its base value
  while its neighbours trend, and the percent bases hold level.

Added

• Regression test pinning the projectionmodels segments of docs Example 10
  (worked-example-contract.md): contract-year loss costs, the
  premium-independent dollar-expense levels (flat fee, trended fee,
  percent-of-claims surcharge), and both contract pins booking to their
  ratio given the page's solved rates.

v0.6.4

[0.6.4] - 2026-07-10

Changed

• The actuarialpy dependency is now an open floor (>=0.41) instead of a
  capped range (>=0.41.0,<0.45.0). The cap made every actuarialpy minor
  release a resolver conflict until a coordinated re-release here; the
  ecosystem policy is now open floors, with the nightly ecosystem smoke
  workflow catching runtime drift. CI keeps a pinned lane at the 0.41.0
  floor edge.