Fix ccl la and nla sign convention

.github/workflows/ci.yml

@@ -32,9 +32,21 @@ jobs:
     needs: linting
     runs-on: ${{ matrix.os }}
     strategy:
+      fail-fast: false
       matrix:
-        os: [ubuntu-latest]
-        python-version: ["3.10", "3.11", "3.12", "3.13"]
+        include:
+          - os: ubuntu-latest
+            python-version: "3.10"
+            tox-env: py310
+          - os: ubuntu-latest
+            python-version: "3.11"
+            tox-env: py311
+          - os: ubuntu-latest
+            python-version: "3.12"
+            tox-env: py312
+          - os: ubuntu-latest
+            python-version: "3.13"
+            tox-env: py313
 
     steps:
       - uses: actions/checkout@v4
@@ -49,7 +61,7 @@ jobs:
         run: uv sync --extra dev
 
       - name: Run tests
-        run: uv run tox -m test
+        run: uv run tox -e ${{ matrix.tox-env }}
 
   finalise:
     runs-on: ubuntu-latest

.gitignore

@@ -5,15 +5,18 @@ __pycache__/
 .pytest_cache/
 .ruff_cache/
 .mypy_cache/
+.pyre/
 .cache/
 .coverage
+.coverage.*
 coverage.xml
 htmlcov/
 
 # Virtual environments
 .venv/
 venv/
 env/
+ENV/
 .tox/
 .nox/
 
@@ -42,6 +45,14 @@ Thumbs.db
 # IDE files
 .vscode/
 .idea/
+*.iml
+
+# PyCharm
+.idea/
+*.iws
+*.ipr
+*.iml
+.idea_modules/
 
 # Local env files
 .env

src/plima/backends/ccl/la.py

@@ -50,12 +50,15 @@ def make_ccl_la_ia_bias(
     validate_greater_than(z_array, threshold=-1.0, name="z")
 
     if amplitude is None:
-        ia_bias = la_amplitude(z_array, a_ia=a_ia)
+        physical_amplitude = la_amplitude(z_array, a_ia=a_ia)
     else:
         amplitude_array = as_finite_float_array(amplitude, name="amplitude")
 
         try:
-            ia_bias = np.broadcast_to(amplitude_array, z_array.shape).astype(
+            physical_amplitude = np.broadcast_to(
+                amplitude_array,
+                z_array.shape,
+            ).astype(
                 np.float64,
                 copy=True,
             )
@@ -66,4 +69,8 @@ def make_ccl_la_ia_bias(
             )
             raise ValueError(msg) from error
 
-    return z_array.astype(np.float64, copy=True), ia_bias.astype(np.float64, copy=True)
+    ia_bias = -physical_amplitude
+
+    return z_array.astype(np.float64, copy=True), ia_bias.astype(
+        np.float64, copy=True
+    )

src/plima/backends/ccl/nla.py

@@ -3,8 +3,10 @@
 This module connects PLIMA NLA amplitude models to the IA bias tuple expected
 by CCL weak lensing tracers.
 
-The NLA amplitude model lives in ``plima.models.nla``. This file only prepares
-the ``(z, ia_bias)`` tuple used by CCL.
+The backend independent NLA model uses a positive physical ``A_IA`` and applies
+the physical NLA response sign internally. CCL weak lensing tracers use the
+opposite sign convention for the IA bias tuple, so this backend flips the sign
+before returning ``(z, ia_bias)``.
 """
 
 from __future__ import annotations
@@ -29,14 +31,22 @@ def make_ccl_nla_ia_bias(
 ) -> tuple[FloatArray, FloatArray]:
     """Return a CCL IA bias tuple for NLA.
 
+    This backend follows the PLIMA user-facing convention that positive
+    ``A_IA`` corresponds to a positive physical NLA amplitude. The sign needed
+    by CCL weak lensing tracers is applied internally, so the returned
+    ``ia_bias`` is ``-amplitude``.
+
     Args:
         z: Redshift values where the IA bias should be sampled.
-        amplitude: Optional precomputed NLA amplitude evaluated at ``z``. If
-            ``None``, the amplitude is computed from ``nla_amplitude``.
-        a_ia: NLA amplitude normalization used when ``amplitude`` is ``None``.
+        amplitude: Optional precomputed positive physical NLA amplitude
+            evaluated at ``z``. If ``None``, the amplitude is computed from
+            ``nla_amplitude``.
+        a_ia: Positive physical NLA amplitude normalization used when
+            ``amplitude`` is ``None``.
 
     Returns:
-        Redshift values and IA bias values sampled on the same grid.
+        Redshift values and CCL sign convention IA bias values sampled on the
+        same grid.
 
     Raises:
         ValueError: If redshifts are not finite, redshifts are outside the
@@ -50,12 +60,15 @@ def make_ccl_nla_ia_bias(
     validate_greater_than(z_array, threshold=-1.0, name="z")
 
     if amplitude is None:
-        ia_bias = nla_amplitude(z_array, a_ia=a_ia)
+        physical_amplitude = nla_amplitude(z_array, a_ia=a_ia)
     else:
         amplitude_array = as_finite_float_array(amplitude, name="amplitude")
 
         try:
-            ia_bias = np.broadcast_to(amplitude_array, z_array.shape).astype(
+            physical_amplitude = np.broadcast_to(
+                amplitude_array,
+                z_array.shape,
+            ).astype(
                 np.float64,
                 copy=True,
             )
@@ -66,4 +79,8 @@ def make_ccl_nla_ia_bias(
             )
             raise ValueError(msg) from error
 
-    return z_array.astype(np.float64, copy=True), ia_bias.astype(np.float64, copy=True)
+    ia_bias = -physical_amplitude
+
+    return z_array.astype(np.float64, copy=True), ia_bias.astype(
+        np.float64, copy=True
+    )