BREOS 0.3.2

[0.3.2] - 2026-06-26

Upgrading: config validation is now strict — a config with an unknown
top-level key (e.g. a typo like batery_kwh) that silently defaulted in
0.3.1 now raises listing the offending key(s). Fix or remove stray keys
before upgrading. All other changes preserve prior behaviour by default.

Removed

• The nrel-pysam runtime dependency. It was only ever reached transitively,
  through pvlib's fit_cec_sam, to fit the CEC single-diode parameters on the
  default PV path. nrel-pysam publishes no Python 3.14 wheel or sdist and was
  the sole blocker to running BREOS on 3.14.

Added

• breos.cec_fit.fit_cec_params: a pure-scipy/pvlib implementation of the
  CEC 6-parameter coefficient calculator (Dobos 2012, DOI:10.1115/1.4005759),
  a drop-in for pvlib.ivtools.sdm.fit_cec_sam. Across every bundled module it
  reproduces the SAM fit to within 0.03% on maximum power over a
  temperature x irradiance grid and 0.004% on annual energy, so model results
  are unchanged. Validated against the nrel-pysam oracle by
  tools/validate_cec_fit.py.
• Python 3.14 support: the 3.14 classifier and CI matrix entry, now that the
  nrel-pysam blocker is gone.
• Config validation now rejects unknown top-level keys. A typo such as
  batery_kwh previously slipped through merge_defaults and silently
  defaulted (e.g. the battery to 0), producing plausible-but-wrong results;
  it now raises listing the offending key(s). The optional montecarlo
  section is recognised so Monte Carlo configs still validate.
• Configurable sky-diffusion (transposition) model via a transposition_model
  config key and --transposition-model / --sky-model CLI flag, threaded
  through calculate_pv_production_dc, the tracking and multi-array variants,
  and the App config surface. Supports isotropic (default), klucher,
  haydavies, reindl, king, perez, and perez-driesse via pvlib's
  get_total_irradiance; the extra inputs the anisotropic models need
  (extraterrestrial DNI, relative airmass) are derived internally. The default
  isotropic reproduces prior results bit-for-bit. Per-array overrides are
  supported in pv_arrays.
• Configurable ground reflectance and Perez coefficients to drive those models
  with real site information: albedo (0-1) or a named surface_type
  ("snow", "sea", "grass", ...) sets the ground-diffuse reflectance for
  every model (previously fixed at pvlib's 0.25), and model_perez selects
  the Perez coefficient set. All three are App config keys with matching
  --albedo / --surface-type / --perez-model CLI flags and per-array
  overrides; not setting them leaves the previous defaults unchanged.

Changed

• The default PV path fits CEC parameters via breos.cec_fit.fit_cec_params
  instead of pvlib.ivtools.sdm.fit_cec_sam; breos/solar.py and the public
  API are otherwise unchanged.
• The two placeholder Generic_400W and Generic_600W_Bifacial catalog
  modules now carry realistic mono-PERC datasheet specifications (their
  previous made-up values fit cleanly under SAM only via an internal
  short-circuit-current heuristic); their nameplate power and keys are
  unchanged.
• resolve_pv_system no longer mutates the merged config in place to record
  the derived n_modules; the resolved count is materialised into a fresh
  dict by resolve_app_config, so the dict wrapped by the frozen
  ResolvedAppConfig is built once and the caller's input dict is left
  untouched.