Changelog

What shipped in each release; why is in the dev diary, what is next in the roadmap.

v0.7.0 — 2026-08-17¶

The fitting subsystem, the domain-contract engine, a hardened vocabulary, and a documentation site rebuilt around them.

Breaking

• The controlled vocabulary is canonical-only — no aliases. Bruker multi-receiver data now loads with the dimension coil (was channels), acquisition dimensions are singular (average, repetition), vocabulary terms are frozen against mutation, and the legacy bruker_group_delay attribute fallback is gone. Rename inbound dimensions with obj.rename({...}) until da.xmr.map_vocab lands. — #65 · #96 · The controlled vocabulary

Added

• da.xmr.fit_amares fits AMARES in the time domain and returns a Dataset of fitted parameters alongside the reconstructed signals. It meets data in either domain and returns it unchanged, normalises away the magnitude scale trap that made Bruker-scale FIDs “converge” on their prior, and writes NaN for a failed voxel rather than a spurious zero. build_prior_knowledge replaces pyAMARES’s positional CSV with a named-peak dict, and the optimizer itself lives behind the optional xmris[fitting] extra, so a bare import xmris never pulls it in. — #67 #69 #70 #80 #81 #82 · #105 · Quick start: fitting · AMARES in depth · diary

• Every domain-sensitive transform now declares the domain it works in, via @computes_in (domain-preserving) or @ensures_domain (funnelling), so passing a FID where a spectrum is wanted converts instead of returning nonsense. — #42 #63 · #73 · #77 · #78 · The Two Domains · Domain contracts in action

• xmris.set_options(auto_convert=False) turns that automatic conversion into a loud error naming the converter to call — strict mode for quantitative work. Global or a context manager, mirroring xr.set_options. — #63 · #79 · The Two Domains

• da.xmr.baseline_als corrects a spectral baseline by asymmetric least squares. — #43 · #44 · Baseline correction

• da.xmr.estimate_group_delay measures a Bruker group delay from the data by minimising the residual, instead of trusting the header value. — #85 · #89 · Bruker — group delay

• da.xmr.widget.scroll_spectra scrolls through a stack of spectra, and da.xmr.widget.apodize tunes line broadening against a live plot. Both render in the built documentation, not only in a running kernel. — #16 · #37 · #38 · #40 · Interactive scrolling · Interactive apodization

Changed

• xmris is released under BSD-3-Clause, replacing AGPL-3.0. — #130

• autophase was rewritten for robustness, and nmrglue is no longer a dependency. — #30 · #36 · Automated phase correction

• Plotting is configured by objects rather than keyword soup — WaterfallConfig, CarpetConfig, PlotTrajectoryConfig, PlotQCGridConfig, reached through da.xmr.plot. — #39 · #41 · Config-based plotting · Waterfall plots · Carpet plots

• apodize_exp, apodize_lg and zero_fill return the representation you handed them, while baseline_als funnels into the spectral domain and stays there. — #63 · #78 · The Two Domains

Fixed

• A bare pip install xmris was unimportable: matplotlib is imported by every import xmris but arrived only transitively through pyAMARES, and vanished when pyAMARES moved to the fitting extra. requires-python also read <=3.13, which PEP 440 resolves to <= 3.13.0 — admitting no 3.13 patch release. Both are fixed, and CI now installs the way a user does. — #122 · #147

• fit_amares reads the canonical reference_frequency attribute, so FIDs produced by simulate_fid fit without a manual mhz=. — #68 · #93

• Waterfall and carpet plots reject complex input instead of silently plotting its real part, and keep the time axis’s units. — #83 · #94

• import xmris no longer emits a DeprecationWarning. — #67 · #92

Documentation

• The site was rebuilt: one directory per chapter, hands-on tutorials split from concept explainers, and a landing page per chapter. — #126 #137 · #139

• The contributor documentation is new in full: The Architecture Contract — the eleven rules every change to src/xmris/ obeys — a page per kind of contribution, and the dev diary. — #72 · #103 · #114 · Contribute · Open a pull request · diary

• Two explainers for the design decisions users hit first: The Two Domains on why a function cares whether it is handed a FID or a spectrum, and The controlled vocabulary on why the names are fixed. — #76 · #100

• The roadmap says what is shipped, in motion, and still being argued about. — #116 · #123 · #145

• Every pull request now publishes a fully executed preview of the site it would produce. — #112 · How the documentation reaches the web · diary

Maintenance

• The documentation pages are the maths tests — every tutorial and explainer is executed by nbmake on both ends of the supported Python range, and a whole-tree docs-style checker gates the merge. — #104 · #110 · diary

• Seven dependency updates, now arriving weekly via Dependabot, plus CI hardening: the site is published from workflow artifacts rather than a 103 MB committed branch, the Codecov uploader comes from PyPI so an outage cannot block every merge, and ruff format is gated for the first time. — #115 #141 #151 · #142 · #146 · #152

Earlier releases¶

v0.1.0 – v0.6.1 predate this changelog. Their contents are the tag list and the commits between them.