Version 0.4.0
=============

**Release Date:** June 2026

**Status:** Stable release

New: relaxation fitting module
------------------------------

A new package, ``epyr.relaxation``, fits time-domain T1/T2 relaxation data,
complementing the existing field-domain lineshape fitting in
``epyr.lineshapes.fitting``.

- ``epyr.relaxation.models``: six decay/recovery functions, ``mono_exponential``,
  ``stretched_exponential``, ``biexponential``, ``inversion_recovery``,
  ``saturation_recovery``, and ``gamma_gaussian_decay`` (a combined
  homogeneous/spectral-diffusion echo-decay model).
- ``epyr.relaxation.fit_relaxation()``: fits a single decay/recovery curve
  with any of the six models, returning a ``RelaxationFitResult`` with
  parameters, errors, R-squared, reduced chi-squared, and an optional
  annotated plot. Mirrors the existing ``fit_epr_signal()`` API.
- ``epyr.relaxation.fit_multiple_decays()``: fits several candidate models
  and ranks them by reduced chi-squared (not R-squared, which is biased
  toward models with more free parameters) to suggest the best-supported
  decay model for a given dataset.
- New example script ``examples/clean/06_relaxation_fitting.py`` demonstrates
  all three entry points on a real T2 echo decay and on synthetic
  bi-exponential and Gamma0/GammaG data.

Relaxation fitting: readable results and rcParams-aware plots
---------------------------------------------------------------

- ``RelaxationFitResult.__repr__`` now returns ``summary()``, so a single
  fit result prints its parameters and errors directly instead of dumping
  the raw dataclass fields (including the full fitted-curve and residual
  arrays).
- ``fit_multiple_decays()`` now returns ``RelaxationFitComparison``, a
  ``dict`` subclass that behaves exactly like the plain dict it replaces
  (subscripting, ``.items()``, iteration) but whose ``__repr__`` prints a
  side-by-side table of R-squared, chi-squared, and every fitted parameter
  across all models.
- Fit plots (``_plot_fit_results``, ``_plot_comparison``) no longer hard-code
  figure size, marker size, line width, or text size. They reuse the
  current figure's size if one is already open, otherwise fall back to
  ``matplotlib.rcParams`` defaults, and likewise read marker/line/font size
  from ``rcParams`` rather than overriding them. The identical change was
  applied to ``epyr.lineshapes.fitting`` for consistency between the two
  fitting modules.

Tutorial notebooks rebuilt
--------------------------

The ``examples/notebooks/`` directory was rebuilt into an eight-notebook
series (``00`` index plus ``01``-``07``) covering loading and plotting,
baseline correction, lineshape fitting, relaxation fitting, signal
processing, FAIR conversion, and physics/units. Each notebook runs top to
bottom on the bundled datasets and is committed without cell outputs.
Superseded notebooks are archived under ``examples/notebooks/_legacy/``.

Bug fix: batch FAIR conversion
------------------------------

``batch_convert_directory()`` forwarded the wrong keyword names to
``convert_bruker_to_fair()`` (``input_file_or_dir`` and ``output_formats``),
raising ``TypeError`` for every file and producing no output. It now passes
``input_file`` and ``formats`` correctly, so batch conversion writes the
requested formats. A regression test covers this.

No backwards-incompatible changes.