v0.4.0

What's Changed

• VisuCoreDiskSliceOrder=disk_reverse_slice_order raises `InvalidData… by @headmeister in #144
• Test suite exercises zero real Bruker data: fixtures root at nonexist… by @headmeister in #145
• JCAMPDX fixes brackets and par anchoring by @headmeister in #146
• 3-D radial / CS fid mis-sized: RADIAL block count omits the parti… by @headmeister in #147
• dim_type desynchronised from the data array (dEPI, SPECTROSCOPY, 2d… by @headmeister in #148
• reordering fix ACQ_obj_order never applied: fid slice axis delive… by @headmeister in #149
• k-space ordering BART reordering, sync docs by @headmeister in #150

Full Changelog: v0.3.0...v0.4.0

Potential breaking changes

This list covers behavior changed during the format-conformance remediation
work. Items marked breaking can alter returned arrays or raise a new,
intentional exception; items marked migration preserve the old API but
should prompt downstream code to move to the explicit replacement.

Data representation

• Migration — PV360 rawdata.jobN: Dataset.data remains the historical
  decoded stream, but now emits FutureWarning because its meaning differs
  from FID .data. Use Dataset.raw for (sample, shot, receiver) data or
  Dataset.kspace for ordered Cartesian k-space.
• New API: Dataset.raw, Dataset.kspace, and
  Dataset.to_kspace(bart=True) provide explicit raw/k-space representations.
  kspace is only available for FIDs and supported Cartesian PV360 jobs.
• Breaking — FIDs: ACQ_obj_order is now applied, so multi-slice FID data
  is returned in object/slice order instead of acquisition order. Code that
  compensated manually must remove its extra permutation.
• Breaking — 2dseq: RECO_transposition / VisuCoreTransposition now
  swaps the first two spatial axes for flagged frames. Historical consumers
  that applied this correction themselves must stop doing so.
• Breaking — 2dseq: VisuCoreDiskSliceOrder=disk_reverse_slice_order is
  applied to the third encoded axis for 3-D data. Earlier releases rejected
  these volumes; code relying on the on-disk slice order must adjust.

Axis and random-access behavior

• Breaking: dim_type is now kept rank-consistent with loaded data,
  including rawdata and traj. Code depending on missing/synthetic labels or
  previous misaligned indices must update its axis selection.
• Breaking: mmap=True is explicitly unsupported for fid, traj, and
  rawdata, and raises UnsupportedDatasetType. It remains supported for
  2dseq; partial encoded-axis slices now return exactly the requested data.

Parsing and dataset discovery

• Behavior change: JCAMP-DX versions are whitespace-normalized and record
  starts are line-anchored. Files previously rejected or crashed may now load.
• Behavior change: passing a PV360 experiment directory now resolves the
  lowest-numbered rawdata.jobN when no fid or 2dseq exists.
• Behavior change: d3proc is loaded as optional metadata for 2dseq and
  participates in legacy word-type/geometry fallback.

Metadata and helpers

• New API: Dataset.frame_group_values returns VisuGroupDepVals
  metadata reshaped to broadcast against 2dseq data. Do not assume legacy
  flat echo-time or diffusion-matrix vectors are the preferred representation.
• New API: Dataset.metadata exposes grouped, snake-case Visu and
  SUBJECT_* metadata. Existing dataset['ParameterName'] access is
  unchanged.
• Breaking: utils.combine_channels now performs RSS as
  sqrt(sum(abs(data)**2)); complex callers that consumed the previous
  complex-square result will receive physically meaningful real magnitudes.

Scope removals and diagnostics

• Breaking: TopSpin ser is no longer advertised by the load sweep and is
  intentionally unsupported.
• Behavior change: direct construction of fid.orig, fid.spiral, and
  fid.navFid produces a companion-specific diagnostic directing callers to
  the parent FID's fid_companions mapping.
• Behavior change: trajectory reports use a fallback filename rather than
  failing when no dataset id is available.