Clarify typed test fixtures

GLOSSARY.md

@@ -1,6 +1,6 @@
-# Firecrown API, Dependency, and Release Language
+# Firecrown Language
 
-Terms used when assessing Firecrown dependencies and the compatibility of releases and pull requests.
+Terms used when assessing Firecrown dependencies, release compatibility, and scientific test expectations.
 
 ## Language
 
@@ -33,3 +33,11 @@ An advisory, technically non-blocking sample of environment rebuildability for a
 
 **API policy verdict**:
 The target-dependent interpretation of a completed static public API comparison: no detected breaking changes, informational breaking changes on `master`, or rejection of detected breaking changes on a support target. The verdict is distinct from whether analysis and reporting succeeded; missing analysis or reporting is not evidence of compatibility, and no detected breaks does not establish complete behavioral compatibility.
+
+## SACC Test Language
+
+**SACC fixture expectation**:
+A reference redshift distribution or angular spectrum retaining the inputs supplied to SACC, or copies of those inputs, for comparison with extracted results.
+
+**Synthetic SACC observation**:
+A supplied test value for a recognized SACC observable, such as an angular power spectrum, angular correlation, or windowed bandpower. It is neither a survey measurement nor necessarily a theory prediction or physically plausible realization; a value with an unknown data-type label has no established observable interpretation.

firecrown/likelihood/_gaussfamily.py

@@ -170,8 +170,8 @@ def __init__(
         self.cholesky: None | npt.NDArray[np.float64] = None
         self.inv_cov: None | npt.NDArray[np.float64] = None
         self.cov_index_map: None | dict[int, int] = None
-        self.theory_vector: None | npt.NDArray[np.double] = None
-        self.data_vector: None | npt.NDArray[np.double] = None
+        self.theory_vector: None | npt.NDArray[np.float64] = None
+        self.data_vector: None | npt.NDArray[np.float64] = None
         self._use_cholesky_for_chisq: bool = use_cholesky
 
     @classmethod

firecrown/likelihood/_gaussian_pointmass.py

@@ -399,10 +399,10 @@ def _prepare_integrand(self, cosmo: pyccl.Cosmology) -> np.ndarray:
 
         # Cosmology calculations
         sc_valid = pyccl.sigma_critical(cosmo, a_lens=aL_all, a_source=aS_all)
-        sc = np.full(mask.shape, np.nan, dtype=float)
+        sc = np.full(mask.shape, np.nan, dtype=np.float64)
         sc[mask] = sc_valid
         D_l_valid = pyccl.angular_diameter_distance(cosmo, a_l)[z_l > 0]
-        D_l = np.full(z_l.shape, np.nan, dtype=float)
+        D_l = np.full(z_l.shape, np.nan, dtype=np.float64)
         D_l[z_l > 0] = D_l_valid
         D_l = D_l[:, None]
         integrand = 1.0 / sc / (D_l * D_l)
@@ -448,7 +448,6 @@ def _build_V(self, betas: np.ndarray) -> np.ndarray:
 
         V = np.zeros((len(row_src_idx), len(lens_tracers)))
         valid = (row_lens_idx >= 0) & (row_src_idx >= 0)
-        beta_rows = np.zeros(valid.sum(), dtype=float)
         beta_rows = betas[row_lens_idx[valid], row_src_idx[valid]]
         V[valid, row_lens_idx[valid]] = beta_rows / (theta[valid] ** 2)
         return V

tests/_types.py

@@ -1,6 +1,10 @@
-"""Shared types for test fixtures."""
+"""Shared types for test fixtures.
 
-from typing import TypeAlias, TypedDict
+Expected arrays retain the inputs supplied to SACC, or copies of those inputs.
+"""
+
+from dataclasses import dataclass
+from typing import TypedDict
 
 from numcosmo_py import Nc
 import numpy as np
@@ -9,22 +13,150 @@
 
 from firecrown.metadata_types import TracerNames
 
-RedshiftDistribution: TypeAlias = tuple[
-    npt.NDArray[np.float64], npt.NDArray[np.float64]
-]
-"""Sampled redshift distribution as (z, dndz), without a normalization guarantee."""
-
-WindowedAngularPowerSpectrum: TypeAlias = tuple[
-    npt.NDArray[np.int64], npt.NDArray[np.float64], sacc.BandpowerWindow
-]
-"""Window ell grid, measured bandpowers, and their bandpower window."""
-
-SaccGalaxyWindowedFixture: TypeAlias = tuple[
-    sacc.Sacc,
-    dict[str, RedshiftDistribution],
-    dict[tuple[TracerNames, str], WindowedAngularPowerSpectrum],
-]
-"""SACC data, distributions by tracer name, and spectra by tracer pair/data type."""
+
+@dataclass(frozen=True, kw_only=True)
+class SampledRedshiftDistribution:
+    """Sampled redshift coordinates and unnormalized number density.
+
+    :param z: Sampled redshift coordinates.
+    :param dndz: Number-density values on the redshift grid; not normalized.
+    """
+
+    z: npt.NDArray[np.float64]
+    dndz: npt.NDArray[np.float64]
+
+
+@dataclass(frozen=True, kw_only=True)
+class SaccGalaxyDistributionFixture:
+    """SACC tracer data and its sampled redshift distributions.
+
+    :param sacc_data: SACC data used by production extraction and construction.
+    :param distributions: Redshift distributions keyed by tracer name.
+    """
+
+    sacc_data: sacc.Sacc
+    distributions: dict[str, SampledRedshiftDistribution]
+
+
+# Harmonic and real-space cross-bin records have distinct types because they
+# represent different physical observables. Their shared fields can evolve
+# independently as fixture values become more physically realistic.
+@dataclass(frozen=True, kw_only=True)
+class SaccHarmonicCrossBinFixture:
+    """Harmonic cross-bin SACC data and expected redshift distributions.
+
+    Both tracer distributions share their original redshift coordinate array,
+    but retain distinct unnormalized number densities. No measurement metadata
+    is inferred here; consumers still exercise production extraction.
+
+    :param sacc_data: Unwindowed harmonic SACC data for the two tracers.
+    :param distributions: Sampled redshift distributions keyed by tracer name.
+    """
+
+    sacc_data: sacc.Sacc
+    distributions: dict[str, SampledRedshiftDistribution]
+
+
+@dataclass(frozen=True, kw_only=True)
+class SaccRealCrossBinFixture:
+    """Real cross-bin SACC data and expected redshift distributions.
+
+    Both tracer distributions share their original redshift coordinate array,
+    but retain distinct unnormalized number densities. No measurement metadata
+    is inferred here; consumers still exercise production extraction.
+
+    :param sacc_data: Unwindowed real-space SACC data for the two tracers.
+    :param distributions: Sampled redshift distributions keyed by tracer name.
+    """
+
+    sacc_data: sacc.Sacc
+    distributions: dict[str, SampledRedshiftDistribution]
+
+
+@dataclass(frozen=True, kw_only=True)
+class HarmonicAngularPowerSpectrum:
+    """Unwindowed harmonic coordinates and synthetic observations.
+
+    :param ells: Integer multipoles for the synthetic observations.
+    :param cells: Synthetic observed values, not survey data or theory predictions.
+    """
+
+    ells: npt.NDArray[np.int64]
+    cells: npt.NDArray[np.float64]
+
+
+@dataclass(frozen=True, kw_only=True)
+class SaccGalaxyHarmonicFixture:
+    """Harmonic SACC data with named expected distributions and spectra.
+
+    :param sacc_data: SACC data used by production extraction and construction.
+    :param distributions: Redshift distributions keyed by tracer name.
+    :param spectra: Synthetic spectra keyed by ordered tracer names and data type.
+    """
+
+    sacc_data: sacc.Sacc
+    distributions: dict[str, SampledRedshiftDistribution]
+    spectra: dict[tuple[TracerNames, str], HarmonicAngularPowerSpectrum]
+
+
+@dataclass(frozen=True, kw_only=True)
+class RealAngularCorrelation:
+    """Unwindowed angular coordinates and synthetic observations.
+
+    :param thetas: Angular coordinates for the observations.
+    :param xis: Synthetic correlations, not survey data or theory predictions.
+    """
+
+    thetas: npt.NDArray[np.float64]
+    xis: npt.NDArray[np.float64]
+
+
+@dataclass(frozen=True, kw_only=True)
+class SaccGalaxyRealFixture:
+    """Real SACC data with named expected distributions and correlations.
+
+    :param sacc_data: SACC data used by production extraction and construction.
+    :param distributions: Redshift distributions keyed by tracer name.
+    :param spectra: Synthetic correlations keyed by ordered tracer names and data type.
+    """
+
+    sacc_data: sacc.Sacc
+    distributions: dict[str, SampledRedshiftDistribution]
+    spectra: dict[tuple[TracerNames, str], RealAngularCorrelation]
+
+
+@dataclass(frozen=True, kw_only=True)
+class WindowedAngularPowerSpectrum:
+    """Window support, synthetic bandpowers, and their shared window.
+
+    Support multipoles need not match the number of bandpowers.
+    SACC observation labels are weighted support sums, not the support grid
+    or normalized band centers.
+
+    :param support_ells: Integer multipoles on which the window is defined.
+    :param bandpowers: Synthetic observed values, not survey data or predictions;
+        one per window column.
+    :param window: SACC BandpowerWindow shared with the fixture's data points;
+        its values are support multipoles and its weights are unnormalized.
+    """
+
+    support_ells: npt.NDArray[np.int64]
+    bandpowers: npt.NDArray[np.float64]
+    window: sacc.BandpowerWindow
+
+
+@dataclass(frozen=True, kw_only=True)
+class SaccGalaxyWindowedFixture:
+    """Windowed SACC data with named expected distributions and spectra.
+
+    :param sacc_data: SACC data used by production extraction and construction.
+    :param distributions: Redshift distributions keyed by tracer name.
+    :param spectra: Windowed spectra keyed by ordered tracer names and data type.
+    """
+
+    sacc_data: sacc.Sacc
+    distributions: dict[str, SampledRedshiftDistribution]
+    spectra: dict[tuple[TracerNames, str], WindowedAngularPowerSpectrum]
 
 
 class NumCosmoCosmologyFixture(TypedDict):