PR 7: adopt CosmoSIS native SACC likelihood (sacc_like), validated against the converter path#255
Draft
cailmdaley wants to merge 43 commits into
Draft
PR 7: adopt CosmoSIS native SACC likelihood (sacc_like), validated against the converter path#255cailmdaley wants to merge 43 commits into
cailmdaley wants to merge 43 commits into
Conversation
PRD #241 row 1. sacc>=2.4 joins core dependencies (sp_validation.sacc_io will be core library code). firecrown v1.15.1 + smokescreen 1.5.6 + an exact pyccl pin form the new [blinding] extra: the blind must be exactly recomputable from the seed at unblinding time, so the theory stack is pinned as a set. firecrown is not on PyPI and hard-depends on numcosmo-py (conda-forge only) plus the cosmosis/cobaya sampler connectors we never import — the new uv-overrides.txt drops those three from resolution; the Dockerfile and README carry the --overrides invocation. Plain installs without the blinding extra are unaffected. requires-python moves 3.11 → 3.12: Smokescreen and firecrown both set a 3.12 floor, and the container base (shapepipe:develop) already runs python:3.12-slim-bookworm — this aligns pyproject with the actual runtime. ruff target-version follows. Validated: uv resolution of '.[test,glass,blinding]' with overrides (235 packages, py3.12) and the full firecrown-core + smokescreen import chain in a fresh venv on candide. The CI image build on this branch is the container-side proof. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019bGhVMAuhTy1gF6DdUc1dQ
Follow-up hardening after empirical validation of the blinding stack on candide. Two facts surfaced that the resolve-only check couldn't see: - firecrown 1.15.1 subclasses npt.NDArray (DataVector); numpy 2.5 made that a non-subclassable typing alias, so firecrown breaks at import on the numpy>=2.0 resolution (2.5.1). The [blinding] extra now carries numpy>=2.2,<2.5 (2.4.3 verified against the full compiled stack + the fast suite), and the Dockerfile requests the bound explicitly to dodge uv #8410 non-movement of an already-installed numpy. - pip-installed firecrown hits NumCosmo (conda-forge-only) at import time through two paths unrelated to cosmic shear: eager re-export of LSST predefined n(z) bins in generators/__init__ (defeating upstream's own lazy __getattr__), and the cluster likelihoods -> lsstdesc-crow -> Ncm.IntegralND C-subclass at module load. scripts/patch_firecrown.py makes the bin re-export lazy, the cluster imports optional, and ships a loud numcosmo_py shim (raises on any real use). Exact-string surgery against pinned v1.15.1, idempotent, fails loudly on a version bump, ends with an import check. Verified: fresh application on a pristine install, idempotent re-run, and the full toy Smokescreen blind end-to-end after patching. CI now also smoke-tests the blinding imports in the built image - the fast suite never imports firecrown, so a broken blinding stack would otherwise ship green. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019bGhVMAuhTy1gF6DdUc1dQ
Add sp_validation.sacc_io: the writer/reader layer for the two-file SACC
layout that becomes the package's standard data-product format.
{version}.sacc analysis vector — NZ tracers, coarse xi+/-, pseudo-Cl
(EE/BB/EB) with a shared BandpowerWindow, COSEBIs,
pure E/B, rho/tau PSF diagnostics; one FullCovariance
assembled block-diagonally (zero cross-blocks).
{version}_xi_fine COSEBIs/pure-EB integration input — same NZ tracers,
fine-grid xi+/-, DiagonalCovariance from TreeCorr
varxip/varxim.
Covariance order is point-insertion order (SACC preserves it bitwise
through FITS). Writers insert in the canonical order — xi+ then xi-, Cl
(ee, bb, eb), COSEBIs (all En then all Bn), pure E/B in _EB_KEYS order
(xip_E, xim_E, xip_B, xim_B, xip_amb, xim_amb, matching
b_modes.calculate_eb_statistics), rho, then tau — but readers never assume
global order: every getter resolves indices through s.indices(dtype,
tracers, **tags). assemble_covariance validates that blocks are contiguous,
ascending and tile the data vector exactly, failing loud otherwise.
Custom data types (pure E/B, rho, tau) all parse under
sacc.parse_data_type_name. Tag filters are plain kwargs; the tags={...}
form silently selects nothing and is never used.
Test suite (test_sacc_io.py, all synthetic and fast): per-writer
round-trips (arrays/tags/windows/NZ bitwise), covariance block alignment
and zero cross-blocks, assemble_covariance failure modes, DiagonalCovariance
round-trip, extract() sub-covariance alignment, a tomographic multi-pair
case, reader/writer mirroring on a mixed file, and the end-to-end two-file
layout. 20 passed.
Co-Authored-By: Claude Opus <noreply@anthropic.com>
Fresh-eyes review caught a correctness bug: readers re-sorted selections by
theta/ell/n, but covariance blocks and bandpower windows stay in insertion
order. On a non-ascending grid the reader output silently desynchronised
from its covariance, and get_pseudo_cl returned sorted cl arrays against
unsorted window columns — internally inconsistent within one return tuple.
Fix by construction, not by sort:
- Writers validate their grids. add_xi/add_pure_eb/add_rho/add_tau require
strictly ascending theta; add_pseudo_cl requires strictly ascending
ell_eff (add_cosebis is inherently safe — it enumerates the mode index).
Out-of-order grids raise a loud ValueError naming the argument.
- Readers drop the sort entirely and return in s.indices (insertion) order,
so every getter is covariance- and window-aligned for ANY file, and
ascending for canonical files. The _sorted_* helpers are replaced by
plain insertion-order accessors (_mean/_tag).
Also:
- _pair normalises (i, j) -> sorted, so get_xi(s, (1, 0)) addresses the same
symmetric shear-shear pair as (0, 1) instead of a silent empty read.
- Module docstring documents the tomographic ξ covariance ordering:
insertion is pair-major ([pair0 xip; pair0 xim; pair1 xip; …]), supplied
to assemble_covariance as one contiguous block matching add_xi call order;
type-major converters (DES 2pt-FITS) permute explicitly via s.indices.
- extract() docstring states tracers takes SACC names, not integer bins.
New tests (26 total, was 20): writers reject non-ascending theta/ell;
(1, 0) == (0, 1) round-trip; a 3-pair tomographic ξ covariance assembled as
one contiguous pair-major block with per-pair sub-blocks recovered via
extract(); get_pseudo_cl window column j <-> ell_eff[j] via window_ind tags.
Co-Authored-By: Claude Opus <noreply@anthropic.com>
Add sacc_writers.py: pure per-statistic *_to_sacc functions (xi coarse/fine,
pseudo-Cl, COSEBIs, pure-E/B, rho/tau) that turn computed arrays into
single-statistic SACC "parts", plus assemble_analysis_sacc, which rebuilds the
single {version}.sacc analysis file with a FullCovariance assembled block-
diagonally in canonical order (per the SACC layout contract — not
concatenate_data_sets, whose BlockDiagonalCovariance the contract rules out).
Add bandpower_window_from_workspace to pseudo_cl.py: the NEW plumbing that
threads NaMaster's get_bandpower_windows() into sacc_io.add_pseudo_cl (the
EE/BB/EB diagonal window is verified identical).
Tests: 12 fast tests, each writer round-tripped and checked against the
contract; a real small-nside NaMaster window round-trip; analysis assembly
proven to cover every point with aligned blocks and zero cross-blocks, incl.
the reload-from-disk DAG path.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…sis_fitting
Add sp_validation.twopoint_convert.sacc_to_twopoint_fits, converting an analysis
SACC into the CosmoSIS 2pt-FITS that 2pt_like (and Sacha's rho/tau
2pt_like_xi_sys fork) reads. Output reproduces today's cosmo_inference
cosmosis_fitting.py assembly HDU-for-HDU and byte-for-byte: NZDATA, XI_PLUS/MINUS,
CELL_EE, the blocked COVMAT (STRT_i offsets) and separate COVMAT_CELL, and the
TAU_{0,2}_PLUS + verbatim RHO_STATS tables.
The data vector and covariance are permuted from SACC's pair-major order to the
2pt-FITS type-major layout via s.indices. The tau covariance is laid in as one
contiguous [tau_0+; tau_2+] block, preserving the tau_0<->tau_2 cross-correlation
that covdat_to_fits carries. Rho/tau HDUs need the rho/tau sidecar FITS: the
RHO_STATS varrho_* variance columns are not stored in the analysis SACC, so the
converter copies them verbatim (never fabricates them); xi, Cl, n(z) and the
covariance are fully reconstructed from SACC alone.
Tests:
- test_twopoint_convert.py: byte-equal vs current cosmosis_fitting.py builders on
deterministic synthetic inputs for all three product shapes (plain xi; xi+Cl;
xi+rho/tau), plus the tau_0<->tau_2 cross-correlation and perturbation teeth.
- test_twopoint_convert_realdata.py (skipif on candide paths): build a SACC from a
real product's own contents, convert, byte-compare vs the current writer.
Observed byte-equal for SP_v1.4.6_leak_corr and a glass_mock sibling; the stale
on-disk files differ only by an extra CELL_BB HDU (older script version) and
~1e-17 float noise on bin edges, documented and guarded.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Add sp_validation.one_covariance_io, the two small pieces of glue between the SACC layout and OneCovariance (which knows nothing of SACC and only exchanges files): - write_nz / nz_table / nz_config_stanza: the source_i NZ tracers of an analysis SACC -> OneCovariance's combined whitespace n(z) file (column 0 = shared z grid, one n(z) column per tomographic bin, no bin edges) plus the [redshift] config stanza pointing at it. Config keys verified against upstream OneCovariance config.ini (zlens_directory/zlens_file/value_loc_in_lensbin); the UNIONS template (cosmo_val/pseudo_cl.py._modify_onecov_config) uses the z_directory alias, exposed via dir_key. Fails fast on missing bins or disagreeing z grids. - covariance_blocks: OneCovariance's flat covariance_list_*.dat output -> dense square block(s) paired with SACC selectors for sacc_io.assemble_covariance, reusing statistics.cov_from_one_covariance for the per-block reshape (col 10 gaussian / col 9 gauss+non-gaussian). Single-block and multi-block (tomography-ready) forms. Tests (test_one_covariance_io.py, 9 passing): reshape to a hand-built matrix + gaussian/gauss+ng column selection + perturbation teeth; blocks feed assemble_covariance and round-trip s.covariance.dense; n(z) write/read round-trip, UNIONS dir_key, header/no-header, and fail-fast on mismatched grids / missing bin / bad value_loc. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The theoretical tau covariance is not (6·nbin)² over [τ+; τ−]. Read from the
write-side (shear_psf_leakage.rho_tau_cov.CovTauTh.build_cov): it is a
(3·nbin)² k-major matrix over {τ0, τ2, τ5} with plus/minus folded into one
component per index — exactly the flavor today's CosmoSIS chain consumes via
covdat_to_fits. rho_tau_to_sacc now scatters that plus-only block into the
τ-plus rows/columns of the 6·nbin τ block (per-k [+;−] insertion order),
leaves τ-minus a vartau diagonal, and keeps plus↔minus cross zero. The kwarg
is renamed tau_cov -> tau_cov_th to name the flavor. tau_cov_th=None keeps a
compact DiagonalCovariance placeholder.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…CC part
calculate_pseudo_cl_map/_catalog already call pseudo_cl_to_sacc_part (from the
prior commit); this defines it — building the EE/BB/EB part via
pseudo_cl_to_sacc(nz, meta, ell_eff, cl_all, wsp) and saving it as the native
pseudo_cl_{ver}.sacc product. Drops the legacy save_pseudo_cl FITS writer (no
external callers) and the unused _NMT_* constants.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…tics mixins
Thin *_to_sacc_part methods that turn each already-computed diagnostic into its
single-statistic SACC part via the born-as-SACC writers:
- cosebis.py: cosebis_to_sacc_part picks the fiducial scale cut's {En,Bn,cov}
(find_conservative_scale_cut_key, else the widest cut — mirroring plot_cosebis)
and writes it; the multi-cut .npz sidecar (the PTE scan) is untouched.
- pure_eb.py: pure_eb_to_sacc_part writes the six PURE_KEYS blocks at
gg.meanr with the results['cov'] block.
- psf_systematics.py: calculate_rho_tau_stats now writes a rho_tau_{base}.sacc
part per version from the handler tables, passing cov_tau_{base}_th.npy as
tau_cov_th when present (the tau-plus CovTauTh inference block), else a
loud diagonal-placeholder fallback.
nz/metadata come from the shared core helpers sacc_nz / sacc_metadata.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…mble)
- run_2pcf.py: after calculate_2pcf, write {ver}_xi_coarse.sacc via
xi_to_sacc(grid='coarse', theta_nom=rnom, npairs, weight) — no covariance
(added at assembly). Drops the save_fits threading (mixin no longer writes
DES-style xi FITS).
- run_2pcf_highres.py: on rank 0, write the terminal {version}_xi_fine.sacc via
xi_to_sacc(grid='fine', variances=[varxip; varxim]) as a DiagonalCovariance.
Deletes write_xi_fits (DES-FITS); keeps the .txt. sacc_io import works on the
bare-host MPI path (no healpy), n(z) read from shear.redshift_path.
- generate_pseudo_cl.py: native product is now pseudo_cl_{ver}.sacc (the mixin
writes it); report reads it back via sacc_io.get_pseudo_cl.
- assemble_sacc.py (new, dual-mode): loads the per-statistic parts in canonical
order and calls assemble_analysis_sacc. Injects the xi-coarse block from the
CosmoCov .txt (already [xi+;xi-]-ordered) and the pseudo-Cl block from the
NaMaster iNKA cov FITS as block-diagonal [EE_EE;BB_BB;EB_EB] (cross-spectrum
blocks dropped, TODO flagged). --allow-placeholder attaches a documented
diagonal for cov-less parts so DAG dry-runs / tests produce a valid
FullCovariance; production requires the real cov inputs (fail-fast otherwise).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ather
Adversarial-review hardening (one HIGH, one MEDIUM, one coverage gap):
- HIGH: n_bins>1 silently truncated — n_bins drove only the NZDATA column
count while data/covariance were read from bin (0, 0), so a 2-bin SACC
emitted a plausible-looking FITS carrying 1/3 of the data. The converter
now fails fast unless n_bins == 1 and the ξ tracer pairs are exactly
{(source_0, source_0)}; tomographic emission lands with the tomographic
round.
- MEDIUM: a ξ-less SACC wrote an empty XI_PLUS and a (0, 0) COVMAT
silently; now a loud ValueError.
- Coverage: every prior covariance test exercised the identity permutation
(single-pair SACC order is already type-major). A (row, col)-encoded
covariance test now pins the exact np.ix_ gather of COVMAT (block-diag
ξ + joint non-adjacent [τ0+; τ2+]) and COVMAT_CELL — any transposition,
offset, or swapped block fails.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016Ko89uZF84ez6HEDcTWbJe
…t_config The migrated pseudo-Cl SACC part calls get_redshift(), which reads cc[version]['shear']['redshift_path']; the synthetic fixture wrote the dndz file but never pointed that key at it (KeyError at core.py:422). Verified: the KeyError layer is resolved; the end-to-end test now fails one layer deeper (test asserts the legacy ELL/EE/BB/EB FITS columns while the migrated writer emits the SACC part - the open wiring seam). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016Ko89uZF84ez6HEDcTWbJe
…-flight Working-tree state of the wiring subagent when it died on the session limit (its final turn also went unreviewed by the safety classifier). Wires cosmo_val.smk/twopoint.smk rules and cv_* scripts to the SACC parts + assemble_sacc rule. NOT dry-run-tested (no snakemake in the shared venv), NOT reviewed; test_calculate_pseudo_cl_catalog_end_to_end still fails at the writer/consumer seam (test reads legacy ELL columns). Next session: verify this diff against the DAG design in the PR-4 plan, reconcile the pseudo-Cl end-to-end test with the born-as-SACC output, add writer-path tests, run the suite, then de-WIP. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016Ko89uZF84ez6HEDcTWbJe
… test calculate_pseudo_cl_catalog is born-as-SACC (pseudo_cl_to_sacc_part writes EE/BB/EB + a shared bandpower window), so the end-to-end test can no longer read legacy ELL/EE/EB/BB FITS columns. Round-trip through sacc_io.get_pseudo_cl instead and assert the bandpower window rides the part per the layout contract. The pinned golden spectra are unchanged (identical computation, SACC serialization); they held bitwise through the round-trip. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01HPfbe2XVTGzbnPn4g7BdN6
… parts
The born-as-SACC parts existed but nothing assembled them. Wire the DAG:
per-statistic parts (xi_coarse, pseudo_cl, cosebis, pure_eb, rho_tau) →
assemble_sacc rule → terminal {version}.sacc.
- rule xi: give the coarse ξ± .sacc output the same reporting-binning wildcards
as the .txt (Snakemake requires one wildcard set per rule; the bare
{version}_xi_coarse.sacc name left the binning wildcards unbound and broke DAG
resolution). run_2pcf.py writes the part to the declared output path.
- rule rho_tau_stats: declare the rho_tau .sacc part (already written by
calculate_rho_tau_stats) as a real output; run_rho_tau.py verifies it.
- new rule assemble_sacc + assemble_sacc_all: load the five parts in canonical
order via assemble_sacc.py, injecting a documented diagonal placeholder for
the cov-less ξ/pseudo-Cℓ blocks (real CosmoCov/NaMaster covariance plugs into
the same --xi-cov/--pseudo-cl-cov seam later — PR-3 converter territory).
- cosmo_val_all: request the terminal {version}.sacc per version.
- cv_pseudo_cl.py docstring: born-as-SACC, no longer FITS.
Dry-runs clean for assemble_sacc_all, cosmo_val_all, bmode_summary.json, and
the fine-grid xi_fine target.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HPfbe2XVTGzbnPn4g7BdN6
… seam
The pure assembler is covered in test_sacc_writers; this pins the DAG-facing
script that loads per-statistic .sacc part *files* and rebuilds {version}.sacc:
canonical block order across all five statistics, the diagonal placeholder for
the cov-less ξ± part, real CosmoCov ξ covariance .txt injection, the fail-loud
path when a required covariance is absent, and the pseudo_cl config toggle.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HPfbe2XVTGzbnPn4g7BdN6
The MPI xi_highres rule shells out to run_2pcf_highres.py by absolute path through the deprecated pure_eb/ compat symlink (it can't use Snakemake's script: directive — mpiexec wraps the python call). Anchor the path on common.py's own location (WORKFLOW_SCRIPTS), which resolves to the generic workflow/scripts of whatever checkout parses the DAG, regardless of which paper composes it — workflow.basedir is unreliable here (under module composition it reflects the composing paper, verified: it resolved to papers/cosmo_val/scripts). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01HPfbe2XVTGzbnPn4g7BdN6
… real cov
The analysis {version}.sacc must be byte-comparable against today's
cosmosis_fitting.py assembly (the SACC layout contract's stated requirement for
PR-3's converter). cosmosis consumes the tagged, blinded pseudo-Cl product
(blind=A, powspace, nbins=32 per config harmonic.fiducial), not the untagged
cv_pseudo_cl diagnostic — so assemble_sacc now inputs the tagged part and
injects its real NaMaster covariance from the matching pseudo_cl_cov FITS
(COVAR_EE_EE/BB_BB/EB_EB → block-diagonal). The untagged cv_pseudo_cl part stays
the cv_summarize_bmodes B-mode diagnostic, unchanged.
The ξ± coarse block keeps a documented diagonal placeholder: its real CosmoCov
covariance is blind/gaussian/mask-keyed in the inference tree, and wiring it
couples cosmo_val to the whole inference covariance DAG — that sourcing is PR-3
converter territory, ready at the --xi-cov seam.
test_assemble_sacc: add the pseudo-Cl COVAR-FITS injection path (the live
default) alongside the existing ξ .txt and placeholder paths.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HPfbe2XVTGzbnPn4g7BdN6
…OFILE test_bmodes_workflow_dry_runs invoked a bare python3.12 (resolves off PATH to intel-python without snakemake) and inherited the login shell's slurm SNAKEMAKE_PROFILE (forces an executor plugin the test env need not have). Use sys.executable (the interpreter pytest/snakemake live in) and drop the profile for the dry-run. The guard now passes and exercises the migrated rule graph (rule xi's coarse-sacc output, the WORKFLOW_SCRIPTS path). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01HPfbe2XVTGzbnPn4g7BdN6
…red to PR 7
The SACC migration removed the xi_plus/xi_minus FITS and switched pseudo_cl to
.sacc, so inference_prep's DAG no longer resolves. It is not reachable from the
cosmo_val suite (cosmo_val_all never requests it), so the cosmo_val DAG stays
clean. Per the PR-4 scope this subsystem is left dormant — a comment block at
the rule head names the stale inputs and states PR 7 (native-SACC inference
consumption) rewires it to consume the assembled {version}.sacc directly,
retiring cosmosis_fitting.py's per-product FITS assembly. Comment-only; no
behavior change.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HPfbe2XVTGzbnPn4g7BdN6
…collision
Both pseudo-Cl rules call calculate_pseudo_cl, which hardcoded the untagged
native pseudo_cl_{ver}.sacc; generate_pseudo_cl.py then renamed it to the tagged
name. In one cosmo_val_all DAG both fire: if the untagged cv_pseudo_cl (blind=
None) ran first, the tagged rule's (blind=A) skip-if-exists silently adopted the
blind=None file and the rename deleted cv_pseudo_cl's declared output — rebuild
loops, and wrong-blind n(z) stamped into the terminal {version}.sacc's pseudo-Cl
part.
Thread out_path through calculate_pseudo_cl so each part is born directly at its
final declared name (tagged for the producer, untagged for the diagnostic). No
shared native basename, no rename; skip-if-exists keys on the declared path, so
the two rules' paths are provably disjoint and the blind that computes each file
matches its name. Multi-version + out_path now raises. generate_pseudo_cl.py and
its CLI updated to pass the declared/native out_path directly.
Regression tests: out_path is honoured (native name untouched) and the
multi-version guard fires.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HPfbe2XVTGzbnPn4g7BdN6
…covariance
{version}.sacc is the terminal science file; shipping it with a var=1.0 diagonal
as the LEADING (ξ±) covariance block — ~20 orders off the real variance — is a
silent catastrophic χ²/PTE for any consumer. Drop the unconditional
placeholder_var=1.0. Default: no real ξ-cov wired → assemble_sacc.py's existing
ValueError fires. The placeholder is now gated behind an explicit opt-in
(cosmo_val.allow_placeholder_cov: true), for dry-run / test configs only. The
pseudo-Cℓ block stays real (pseudo_cl_cov input); the integration tests pass
placeholder_var explicitly and are unaffected.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HPfbe2XVTGzbnPn4g7BdN6
test_bmodes_workflow_dry_runs uses papers/bmodes, whose config has no cosmo_val block, so cosmo_val.smk (the born-as-SACC + assemble rules) is never included — the guard couldn't catch a break in them. Add test_cosmo_val_workflow_assemble_ dry_runs targeting assemble_sacc_all in papers/cosmo_val (the only paper that includes cosmo_val.smk): asserts the DAG resolves and each assemble_sacc job pulls the tagged, blinded pseudo-Cl part + its NaMaster cov (not the untagged diagnostic) plus all five per-statistic parts. Shared _dry_run helper factored out of the bmodes guard. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01HPfbe2XVTGzbnPn4g7BdN6
…f call
run_xi_sweep.py called run_2pcf(..., save_fits=True) — a kwarg the SACC
migration removed, so every invocation TypeErrors. The sweep consumes only the
.txt dump; drop the kwarg. run_2pcf is born-as-SACC, so give its coarse part a
grid-qualified sacc_out (the default {ver}_xi_coarse.sacc carries no binning, so
the reporting + integration grids would collide per version). Stale "+ ξ+/ξ-
FITS" docstring dropped.
test_cli_seams: bind the sweep's exact run_2pcf call against the live signature
(and assert save_fits no longer binds) so this CLI seam can't silently rot —
the compute is cluster-only and never exercised by the fast suite.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HPfbe2XVTGzbnPn4g7BdN6
_from_snakemake built part_paths with a hasattr filter, so a typo'd input
keyword (cosebi for cosebis) silently omitted that statistic from the terminal
{version}.sacc — the exact silent-truncation class this series has been bitten
by. assemble_sacc now takes an `expected` list (the statistics the caller wired,
from its config toggles) and raises if any is missing from part_paths or names a
non-CANONICAL statistic. The rule derives expected from cv_assemble_inputs so it
tracks the include_pseudo_cl toggle. CLI path is already typo-safe (argparse
rejects unknown flags), so it passes expected=None.
Tests: typo'd input key raises; typo in the expected list itself raises.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HPfbe2XVTGzbnPn4g7BdN6
Adopt CosmoSIS's native SACC likelihood for the ξ± inference path through a thin
sp_validation-owned subclass that fixes two upstream defects:
1. arcmin→rad theta-tag conversion — upstream sacc_like builds its theory
spline on block[section,"theta"] (radians) but evaluates it at raw SACC
theta tags (arcmin), silently returning theory≈0 and collapsing χ². 2pt_like
converts to radians for exactly this reason; sacc_like never does.
2. an ordering guard — the data vector (get_mean, insertion order) and the
theory loop (get_data_types × tracer_combinations × points) agree only for
type-major files; a pair-major tomographic file would silently misalign.
The guard reconstructs the theory-loop order and requires it == arange.
Factory pattern (setup imports the upstream class from csl_dir and subclasses at
call time) so importing the module never needs cosmosis; not imported by
__init__. build_data calls super() first (scale cuts run in arcmin, matching
2pt_like's angle_range grammar) then converts + guards.
Tests (cosmosis + CSL_DIR gated, skip cleanly otherwise) prove in-process
equality against 2pt_like on the PR-3 converter FITS: machine-zero equality on
synthetic (χ²=1.18829133107, Δχ²=0) and real data (χ²=420310.753739, Δχ²=0, N=40),
the unit-gap tripwire (raw sacc_like 631× off), scale-cut equivalence, identical
perturbation response, the ordering guard, and real-type-scoped theta conversion.
pyproject: cosmosis>=3.25 extra (inert in CI, which installs [test,glass,blinding]).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EnV8NPWhkxS2SxGVgGSJyt
Add the CosmoSIS pipeline ini template for the native SACC path (cosmosis_pipeline_A_ia_sacc.ini): copies the A_ia template, swapping load_nz_fits→load_nz_sacc (native n(z) from the SACC's source_i NZ tracers) and 2pt_like→sacc_like (the shim, via SP_VALIDATION_MODULES, csl_dir=COSMOSIS_DIR, like_name=2pt_like for block-key parity). Same numeric scale cuts as A_ia, in the sacc_like angle_range grammar (full data-type names + source_0 pairs). generate_inference_config.py fills a template's [DEFAULT] section with concrete paths (SCRATCH, FITS_FILE|SACC_FILE, COSMOSIS_DIR, SP_VALIDATION_MODULES resolved from sp_validation.__file__). Dual-mode (snakemake object OR argparse CLI) like assemble_sacc.py; plain text processing (the pipeline.sh sed idiom) that REPLACES an existing DEFAULT key in place and prepends only new ones — no configparser round-trip (which would strip comments + %(...)s interpolation) and no duplicate DEFAULT key (which CosmoSIS's strict parser rejects). Tests (no cosmosis needed): both templates fill + interpolate cleanly, the referenced module file exists, no %(...)s placeholder survives, missing-DEFAULT raises. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01EnV8NPWhkxS2SxGVgGSJyt
Revive the real-data inference_prep from its dormant, pre-SACC state. It now
consumes the assembled analysis {version}.sacc (cosmo_val.smk's assemble_sacc,
bound lazily through cv_analysis_sacc) and emits the A_ia (IA-only, ξ±) file-prep
products:
(a) the converter 2pt-FITS (sacc_to_twopoint_fits, pure ξ — no rho/tau
sidecars, A_ia scope) + a generated 2pt_like ini (the validating/legacy
path), and
(b) a generated sacc_like ini pointing at the SACC (the native path, validated
bit-for-bit against (a)).
The cosmosis_fitting.py real-data assembly is retired from this rule; the
glass-mock rules keep it (their SACC migration is out of scope, noted inline).
inference_fiducial extends to all three prep outputs.
CSL_DIR is read lazily (inference.smk is parsed by every paper workflow, but only
cosmo_val carries inference.csl_dir) — a missing key still fails loudly, just at
DAG time rather than at parse time for an unrelated (bmodes) workflow.
config.yaml: add inference.csl_dir. The workflow dry-run guard gains a test
covering the revived inference_prep DAG (SACC in, converter FITS + both engine
inis out, no cosmosis_fitting.py real-data assembly).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EnV8NPWhkxS2SxGVgGSJyt
…rcmin tags Review finding (MEDIUM): the shim converted theta tags arcmin→rad in place on self.sacc_data. Upstream do_likelihood's save_theory / save_realization paths copy self.sacc_data and overwrite only point values, so they were writing radian theta tags into the saved SACC — any arcmin-assuming consumer (sacc_io.get_xi, or re-ingesting the file as data_file, which double-converts to ~8.5e-8) is then silently off by 3437×. Keep self.sacc_data in arcmin always. build_data now builds a separate self._sacc_data_rad copy with the real-category theta tags converted, and extract_theory_points swaps it in around super().extract_theory_points(block) in a try/finally — so the theory spline sees radians while everything after (save_theory / save_realization) sees the untouched arcmin original. The ordering guard runs on the arcmin object (ordering is unit-independent). Theory equality is unchanged (the swap produces identical results): synthetic χ²=1.18829133107, real-data χ²=420310.753739, both Δχ²=0 vs 2pt_like. New test: run with save_theory set, reload, assert the saved theta tags equal the input file's arcmin tags exactly (and are NOT the radian conversion) and the saved values equal the theory vector; assert a second execute() yields the identical χ² (guards against a double-conversion regression). The theta-scoping test now checks the rad copy carries radians AND the original stays arcmin. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01EnV8NPWhkxS2SxGVgGSJyt
Review finding (MEDIUM): the two pipeline ini templates were referenced as params, so there was no DAG edge — editing a template never regenerated the per-version config files. Bind both as rule inputs (referenced via input.* in the run block). The templates are source files, so anchor them on the running checkout (INFERENCE_TEMPLATE_DIR, off the workflow dir's parent) rather than the env-overridable COSMO_INFERENCE output root; the generated per-version configs still land in COSMO_INFERENCE (INFERENCE_CONFIG_OUT). In a normal run the two roots coincide; the split is what lets a template edit in this checkout drive the DAG. The dry-run guard now asserts both templates appear as inference_prep inputs. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01EnV8NPWhkxS2SxGVgGSJyt
…cosmosis CI (test_imports.py::test_package_module_imports[sp_validation.sacc_like_unions]) failed: the CI image has the science stack but no cosmosis, and the shim imported `from cosmosis.datablock import SectionOptions, option_section` at module level. cosmosis is an optional dependency, so a bare `import sp_validation.sacc_like_unions` must succeed without it (test_imports bare-imports every top-level module). Move the cosmosis import into setup() (call time). Nothing at top level touches cosmosis now — only os/sys/numpy (all in the image). The factory's subclass of the upstream SaccClLikelihood already deferred (setup imports it from csl_dir), so setup() is the single cosmosis entry point. Verified: in the container (numpy present, cosmosis genuinely absent — the CI condition), `import sp_validation.sacc_like_unions` succeeds with setup/execute/ cleanup/ARCMIN_TO_RAD all present, and sacc_like_unions is absent from the import-sweep failures. Full function intact: test_sacc_like.py 8 passed under the venv (cosmosis present). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01EnV8NPWhkxS2SxGVgGSJyt
Adds twopoint_convert (SACC → 2pt-FITS, byte-compatible with cosmosis_fitting) and one_covariance_io (SACC ↔ OneCovariance glue), with their test suites. Rebuilt on feat/sacc-2-sacc-io so the diff is converter-only; adapted to that branch's canonical sacc_io — ξ reads use grid='reporting' (was 'coarse'). Optional-block probing (pseudo-Cℓ, τ) stays on raw s.indices behind get_data_types/use_rho_tau guards, so the new fail-loud sacc_io readers don't trip on absent blocks. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WzUt7VbtXwr2SCHUdiQTyt
cailmdaley
force-pushed
the
feat/sacc-4-cosmo-val-sacc
branch
from
July 16, 2026 09:50
149a50c to
72e680f
Compare
Consistency follows tagging semantics: the grid tag declares which binning a set of points lives on, so all same-length theta arrays under one tag value must be bitwise identical (sacc never validates angles, and grids diverging at floating-point level choke CosmoSIS downstream). Different lengths within a tag group pass (scale-cut subsets); grids under different tag values are unconstrained (reporting vs integration differ by design). The rho/tau/pure-EB writers now tag their points grid="reporting" by default (overridable via grid=), so they join xi's consistency group and no untagged group appears in our own files. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MN9VazXKHUHQg16kiG7Ufk
…seudo-Cl grid="reporting" The theta consistency guard generalizes to both angular domains: theta and ell points are grouped separately by grid tag value, and within a tag value all same-length grids must be bitwise identical. Two ell series sharing a grid must also carry equal bandpower windows (window ells and weight matrix); series without windows skip that check. add_pseudo_cl now stamps grid="reporting" on every point by default (overridable), joining the merge guard's consistency groups; since sacc's add_ell_cl accepts no extra tags, its per-point insertion (ell + shared window + window_ind) is inlined. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MN9VazXKHUHQg16kiG7Ufk
…acc-3-2pt-converter
…ty selections Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MN9VazXKHUHQg16kiG7Ufk
One module whose job is converting between the SACC product and external analysis-tool file formats (CosmoSIS 2pt-FITS, OneCovariance). Public API unchanged; sacc_io untouched. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01766vawzi2XqrgoyHmeHEY9
Cail's ruling: a single sacc_io.py carries the format contract and the external-tool converters (CosmoSIS 2pt-FITS, OneCovariance). Public API unchanged; the sacc_io import name every downstream branch uses is untouched. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01766vawzi2XqrgoyHmeHEY9
…feat/sacc-7-sacc-like # Conflicts: # Dockerfile # pyproject.toml # src/sp_validation/sacc_io.py # src/sp_validation/tests/test_sacc_io.py # workflow/rules/inference.smk Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01766vawzi2XqrgoyHmeHEY9
The merged layer order ran the [blinding] editable install before COPY ., so uv-overrides.txt (and the source tree) did not exist in the build context. The blinding install is itself 'uv pip install -e .[blinding]', so the old final --no-deps editable layer is redundant and is dropped. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01766vawzi2XqrgoyHmeHEY9
The feat/sacc-3 merge brought the guard-era sacc_io whose save() now
requires a keyword-only `type={'data','mock'}` provenance tag (PRD #241
§4) and whose load() refuses unblinded `type='data'`. Application code and
tests from the pr4/pr7 lineage still called the old `save(s, path)` and
referenced a template file the #236 cleanup had relocated, leaving 21 CI
failures outside the merge-conflict set. This adapts the callers to the
reviewed contract (sacc_io is unchanged).
save() type= threading:
- CosmologyValidation gains a `sacc_type` ctor kwarg (default "data"),
stamped by every part-writer: cosebis/pure_eb/psf_systematics/pseudo_cl
now `save(..., type=self.sacc_type)`.
- assemble_sacc inherits provenance from its parts: `type=metadata["type"]`.
- run_2pcf uses `cv.sacc_type`; run_2pcf_highres stamps "data" (real-catalogue
fine covariance).
- Synthetic-data tests adopt the reviewed idiom `type="mock"` (loads freely):
test_sacc_writers _roundtrip + reload helper, test_assemble_sacc part writer,
and the test_pseudo_cl fixture (`sacc_type="mock"`).
Relocated FITS engine template:
- The #236 folder cleanup moved cosmosis_pipeline_A_ia.ini into
cosmosis_config/templates/ (still used by the legacy cosmosis_fitting.py
path). The new pr7 Snakemake regime — inference.smk's INFERENCE_TEMPLATE_DIR,
generate_inference_config, and its tests — resolves templates from
cosmosis_config/ directly, where pr7 placed A_ia_sacc.ini but never copied
its FITS sibling. Restore A_ia.ini alongside the sacc template.
Fast suite: 254 passed, 1 skipped. Two remaining failures
(test_calculate_pure_eb pinned-value drift; test_configured_paths_exist_on_candide)
are pre-existing and environmental — they fail identically with these changes
stashed, and neither is in the API-drift set.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01766vawzi2XqrgoyHmeHEY9
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #254. Row 7 of PRD #241 — the final PR of the SACC/Smokescreen series: the inference pipeline consumes the assembled
{version}.saccnatively through CosmoSIS'slikelihood/sacc/sacc_like.py, with the PR-3 converter path (SACC → 2pt-FITS →2pt_like) retained as the cross-validating twin. Each path validates the other, per PRD §2.The "prototype-grade" verdict, grounded
The PRD flagged
sacc_like's ξ± path as prototype-grade. A source read of CSL @4fd2f1cfound the concrete mechanisms:sacc_likelihoods/twopoint.pybuilds the theory spline onblock[section, "theta"](radians, CosmoSIS convention) and evaluates it at raw SACCthetatags (arcmin, SACC/firecrown convention).2pt_likeexplicitly converts everything to radians (2pt_like.pyL179–183);sacc_likenever does — the spline is evaluated ~3437× outside its grid,SpectrumInterpreturns 0 there, and χ² silently collapses to dᵀC⁻¹d. Its only production use (HSC Y3) was the ℓ-space Cℓ path, which is unit-free — hence never caught.get_mean()(insertion order); theory is a type-major loop. A comment claimsto_canonical_orderwas called on load — it never is. Single-pair files align by construction; tomographic pair-major files would silently misalign.sacc_like.pyL106 references undefinedt1, t2in thekeep_tracersremoval path.)Adoption = subclass shim, not fork
src/sp_validation/sacc_like_unions.pysubclasses upstreamSaccClLikelihood(factory-loaded fromcsl_dir), containing exactly two fixes: theory extraction runs against a radian-θ copy of the loaded SACC (self.sacc_datastays arcmin, so scale cuts keep arcmin ini ergonomics and upstream'ssave_theory/save_realizationnever write radian tags), and a loud ordering guard that reconstructs the upstream theory-loop order and requires it to match insertion order. Everything else — scale-cut grammar, Sellentin/Hartlap, windows for Cℓ — rides upstream unmodified. n(z) loads natively via CSL'snumber_density/load_nz_sacc.like_name = 2pt_likekeeps every block key identical across engines, so chain post-processing is engine-agnostic.Validation (all observed, in-process, no sampler/CAMB needed)
Both likelihood modules run standalone against identical synthetic theory DataBlocks:
2pt_like-on-converter-FITS: synthetic (realistic covariance) χ² = 1.18829133107 both, Δχ² = 0, max|Δtheory| = 0; real data (SP_v1.4.6_leak_corr, ξ-only, N=40) χ² = 420310.753739 both, Δχ² = 0.sacc_likeon the same arcmin SACC is ~631× off — this test failing one day means upstream fixed units and the shim should be retired.n, spectrumelluntouched);save_theoryoutput verified to carry arcmin tags and theory values, with re-execution χ²-stable.Workflow revival
inference_prephonors PR 4's dormant-note contract: input = the assembled{version}.sacc(+ the two ini templates, DAG-tracked), outputs = converter 2pt-FITS + generated2pt_likeini + generatedsacc_likeini (workflow/scripts/generate_inference_config.py, dual-mode likeassemble_sacc.py).cosmosis_fitting.pyis retired from the real-data path; the glass-mock rules keep it, and the PSF/xi_sys pipeline variant stays on2pt_like(both flagged in comments — follow-up work, not silent scope creep).snakemake -n inference_fiducialresolves the full chain: parts →assemble_sacc→inference_prep→inference_fiducial.Environment
New inert
[cosmosis]extra (cosmosis>=3.25; pip-installs cleanly on py3.12). Tests gate onimportorskip("cosmosis")+ aCSL_DIRenv var pointing at a CSL checkout — they skip in CI (image carries neither) and run on candide. CI runs the rest of the suite as usual.Adversarial review
Fresh-context multi-lens review (5 lenses, 3 refuters per finding) before this draft opened; two MEDIUM findings confirmed and fixed: (1)
save_theory/save_realizationwrote radian θ tags into saved SACCs (in-place mutation leak — now a swap-in copy, regression-tested); (2) the ini templates were ruleparamsrather thaninputs, so template edits didn't propagate to generated configs (now DAG edges, asserted by the dry-run guard).Stack note
Base =
feat/sacc-4-cosmo-val-sacc(PR #251), withfeat/sacc-3-2pt-converter(PR #249) merged in — this PR needs both PR 4's workflow surface and PR 3's converter, so its diff shows #249's files until the stack collapses onto develop. Net-new in THIS PR:sacc_like_unions.py,test_sacc_like.py,test_generate_inference_config.py,cosmosis_pipeline_A_ia_sacc.ini,generate_inference_config.py; changed:inference.smk,config.yaml,pyproject.toml,test_bmodes_workflow_dry_run.py.Recommended follow-ups (team call): file the unit gap + the L106
keep_tracersbug upstream at joezuntz/cosmosis-standard-library; migrate the PSF/xi_sys variant and glass-mock prep; tomographic support in both the converter and the shim's guard-relaxation.— Claude (Fable) on behalf of Cail.
🤖 Generated with Claude Code
https://claude.ai/code/session_01EnV8NPWhkxS2SxGVgGSJyt