Merge the tomography branch into develop #375

Activity

1. cailmdaley

   added this to the Tomography branch merged into develop milestone on Oct 2, 2026
2. cailmdaley

   added a parent issue on Oct 2, 2026

   • sp_validation: extend to tomography #205
3. cailmdaley

   added sub-issues on Oct 2, 2026

   • Merge develop into the tomography branch #374
   • Tomographic SACC writers: one tracer and n(z) per bin #376
4. LisaGoh commented on Oct 20, 2026

   

   LisaGoh

   on Oct 20, 2026

   Member

   More actions

   With the previous merges into develop and the work on defining the sacc output format, #256 and #306 have become quite messy and their scopes overlap with some of what has already been done in develop. I ran Claude to try to disentangle everything, and this was what we worked out together:

   Step 5: how #256 and #306 split, and the path to tomographic real-space 2PCFs in SACC

   The two PRs are split by product: #256 computes the covariance, #306 builds the data vector and writes both into SACC. They only meet through a shared naming and bin-order contract.

   	#256 (covariance)	#306 (data vector + SACC)
   Produces	OneCovariance ξ± matrix .dat (CosmoCov removed)	one SACC file with ξ± per bin pair, carrying the OneCovariance covariance
   Files	rules/covariance.smk, common.py	real_space.py, run_2pcf.py, rules/twopoint.smk, sacc_writers.py / assemble_sacc.py
   Tomography	{tomo_suffix} = _tomo	{tomo_suffix} = _tomo
   State	develop merged in; dry runs pass	to merge develop, then rescope to tomographic SACC (#376)

   #306 takes on the OneCovariance → SACC step. #256 stops at the dense .dat matrix. #306 extends xi_to_sacc to tomographic bins, and also reads the OneCovariance matrix into the SACC covariance block in the same bin-pair order. It replaces the jackknife/shot-noise estimate in the analysis file, as assemble_sacc does today for the non-tomographic case.

   The contract between the two

   • Same _tomo suffix in both file names.
   • Bin-pair order (1,1), (1,2), …, (n,n) (itertools.combinations_with_replacement over bin IDs > 0), ξ+ then ξ−, by θ, in both the SACC data points and the covariance.
   • Per-bin n(z) from the same multi-column file: the catalogue entry's shear.redshift_path.

   Workflow once both land

   1. Catalogue entry in cat_config.yaml:
      • shear.tomo_bin_col, a multi-column shear.redshift_path, per-bin cov_th values;
      • blind: set.
   2. Data vector: rule xi → run_2pcf.py → calculate_2pcf_version(compute_tomography=True) measures every bin pair with TreeCorr → …_npatch=…_tomo.sacc.
   3. Covariance: covariance_tomo → covariance_ini writes the OneCovariance .ini → covariance_run → covariance_…_tomo.dat.
   4. SACC (Feature/305 adapt tomographic 2pcf script to sacc io format #306): the OneCovariance matrix goes into the SACC covariance block, and the part is sealed under the catalogue's blind.
   5. Inference reads the single SACC file.

   Merge order

   1. Add tomographic and 3x2pt covariance realspace measurement #256: covariance (OneCovariance replaces CosmoCov). It also adds the tomo_suffix / npatch wildcard constraints both PRs use.
   2. Feature/305 adapt tomographic 2pcf script to sacc io format #306: merge develop; tomographic ξ± → SACC, extending xi_to_sacc (Tomographic SACC writers: one tracer and n(z) per bin #376), plus the OneCovariance .dat → SACC covariance step.
   3. 3x2pt (γ_t, w(θ) and their covariance): deferred until there is a lens sample; separate issue.

   ---

   TO-DO: we can open a separate issue to tackle the 3x2pt part!