v2026.02.2

Change the default Python version (in the environment.yaml file) to Python 3.13

Full Changelog: v2026.02.1...v2026.02.2

v2026.01.1

• Created DEMO.ipynb notebook to provide example input files, illustrate their formats, show how to run the code from a script, and how to load and manipulate the outputs.
• The config.yaml has been reorganised with clearer names for some of the options.
• Add interpolation method choice for galaxy and magnification bias

Full Changelog: v2026.02.0...v2026.02.1

v2026.02.0

Jaxified harmonic-space Gaussian covariance: now the code is ~10 times faster and can compute the 1/fsky Gaussian covariance over all integer ells in ~100s for 13 redshift bins and ell_max = 5000.

Full Changelog: v2026.01.2...v2026.02.0

v2026.01.2

The harmonic-space 1/fsky Gaussian covariance has been updated to provide higher numerical precision.

Previously, the covariance was calculated using the value $C_\ell$ at the bin center. The updated version computes the $C_\ell$ for every integer $\ell$ within falling within an $\ell$ bin and calculates their average. This approach is more accurate, as it doesn't assume the $C_\ell$ to be slowly-varying in the bin width.

NOTE: as the new, more precise binning is the default, this change will break compatibility with previous versions. To reproduce the old results, set in the config file

precision:
  cov_hs_g_ell_bin_average: False

Full Changelog: v2026.01.1...v2026.01.2

v2026.01.1

Small bugfix: cloelib was imported even when not requiring the COSEBIs covariance. Fixed with a lazy import.

Full Changelog: v2026.01.0...v2026.01.1

v2026.01.0

This release introduces the $(E_n, B_n)$ (gaussian) COSEBIs covariance.
⚠️ Note: To compute this covariance, the code now requires cloelib as an additional dependency for calculating the COSEBIs filter functions.

Additional Changes:

• [Refactoring]: Implemented a CovarianceProjector parent class for CovRealSpace and CovCOSEBIs. This centralizes common computations and streamlines the individual classes.
• [Bugfixes]: Fixed a bug in the Shot Noise (SN) term of the 2PCF covariance that caused errors when ngal_lens != ngal_source.
• [Testing]: Implemented a more comprehensive regression test suite to ensure long-term stability and prevent future breaking changes.

Full Changelog: v2025.12.0...v2026.01.0

v2025.12.0

This release features the largest internal code restructuring to date. Covariance arrays are now stored in objects of the FrozenDict class - which inherits from the dict class while overriding several methods to prevent accidental addition, deletion and modification of items. This allows for a much tidier and streamlined handling of the covariance arrays for the different statistics, terms, probes and dimensions. The change does not impact the end user, but it lays the groundwork for more complex interfaces and general code development going forward.

• covariance arrays are now stored in pre-initialised FrozenDict objects
• ~50% faster partial-sky computation, exploiting zpairs symmetries
• improvemed and unified spin2 and spin0 functions for Namaster partial-sky covariance
• simplified covariance reshaping routines

v2025.11.0

• Speed up the coupled Gaussian covariance (with namaster) case
• Tidy up the binning configs
• Minor bugfix: lens n(z) were normalized twice after the shift
• More informative runtime prints

Full Changelog: v2025.10.1...v2025.11.0

v2025.10.1

Minor bugfixes

Full Changelog: v2025.10.0...v2025.10.1

v2025.10.0

• Minor bugfixes for partial-sky computation with input Cls
• Added heracles and euclidlib as optional dependencies in environment.yaml
• Implement euclidlib/heracles/cloelike-compatible .fits format for the output covariance (thanks to @JaimeRZP for the help!)

Full Changelog: v2025.09.3...v2025.10.0