test: mass-sensitivity asserts for jax_likelihood median literals

scripts/imaging/jax_likelihood/mge_group.py

@@ -162,7 +162,16 @@
     profile_list=bulge_gaussian_list,
 )
 
+# The main mass is anchored near the `simulator/simple.py` truth (einstein_radius=1.6) with a
+# prior whose median is that truth. At the config-default prior median (einstein_radius=4.0) the
+# positive-only solver zeroes the source's solved intensities and the vmap literal below is
+# bit-identical for ANY source-plane mass structure (audit 2026-08-06, autolens_workspace_test#253)
+# — the mass-sensitivity assertion after the literal only works on a retained source.
+
 mass = af.Model(al.mp.Isothermal)
+mass.centre = (0.0, 0.0)
+mass.ell_comps = al.convert.ell_comps_from(axis_ratio=0.8, angle=45.0)
+mass.einstein_radius = af.UniformPrior(lower_limit=1.1, upper_limit=2.1)
 
 shear = af.Model(al.mp.ExternalShear)
 
@@ -313,12 +322,46 @@
 
 np.testing.assert_allclose(
     np.array(result),
-    -29060.21511937,
+    -28830.547173,
     rtol=1e-4,
     err_msg="mge_group: JAX vmap likelihood mismatch",
 )
 
 
+"""
+__Mass Sensitivity__
+
+The literal above is evaluated at the model's prior medians, where a +5% change of
+every lens mass parameter moves this likelihood by less than the literal's rtol
+(audit 2026-08-06, autolens_workspace_test#253) — the literal alone would pass a
+source-plane mass regression. This block pins mass sensitivity directly; the floor
+is the audit-measured response divided by five (margin for platform drift).
+"""
+mass_indices = [
+    i
+    for i, name in enumerate(model.model_component_and_parameter_names)
+    if ".mass." in name
+    and "centre" not in name
+    and "ell_comps" not in name
+    and "redshift" not in name
+]
+assert mass_indices, "imaging/mge_group: no mass parameters found for sensitivity check"
+
+parameters_perturbed = np.array(model.physical_values_from_prior_medians)
+for i in mass_indices:
+    parameters_perturbed[i] *= 1.05
+
+ll_median = float(np.asarray(result).ravel()[0])
+ll_perturbed = float(
+    np.asarray(fitness._vmap(jnp.array(parameters_perturbed[None, :]))).ravel()[0]
+)
+assert abs(ll_perturbed - ll_median) > 9.0, (
+    f"imaging/mge_group: likelihood insensitive to a +5% lens-mass perturbation "
+    f"(median={ll_median}, perturbed={ll_perturbed}) — source-plane mass pipeline regression?"
+)
+print("PASS: mass-sensitivity floor exceeded.")
+
+
 """
 __Path A: jit-wrap ``analysis.fit_from``__
 """

scripts/interferometer/datacube/delaunay.py

@@ -257,6 +257,40 @@
 )
 
 
+"""
+__Mass Sensitivity__
+
+The literal above is evaluated at the model's prior medians, where a +5% change of
+every lens mass parameter moves this likelihood by less than the literal's rtol
+(audit 2026-08-06, autolens_workspace_test#253) — the literal alone would pass a
+source-plane mass regression. This block pins mass sensitivity directly; the floor
+is the audit-measured response divided by five (margin for platform drift).
+"""
+mass_indices = [
+    i
+    for i, name in enumerate(model.model_component_and_parameter_names)
+    if ".mass." in name
+    and "centre" not in name
+    and "ell_comps" not in name
+    and "redshift" not in name
+]
+assert mass_indices, "interferometer/datacube/delaunay: no mass parameters found for sensitivity check"
+
+parameters_perturbed = np.array(model.physical_values_from_prior_medians)
+for i in mass_indices:
+    parameters_perturbed[i] *= 1.05
+
+ll_median = float(np.asarray(result).ravel()[0])
+ll_perturbed = float(
+    np.asarray(fitness._vmap(jnp.array(parameters_perturbed[None, :]))).ravel()[0]
+)
+assert abs(ll_perturbed - ll_median) > 0.1, (
+    f"interferometer/datacube/delaunay: likelihood insensitive to a +5% lens-mass perturbation "
+    f"(median={ll_median}, perturbed={ll_perturbed}) — source-plane mass pipeline regression?"
+)
+print("PASS: mass-sensitivity floor exceeded.")
+
+
 """
 __Path A: jit-wrap parameter-vector entry point__
 

scripts/interferometer/datacube/rectangular.py

@@ -239,6 +239,40 @@
 )
 
 
+"""
+__Mass Sensitivity__
+
+The literal above is evaluated at the model's prior medians, where a +5% change of
+every lens mass parameter moves this likelihood by less than the literal's rtol
+(audit 2026-08-06, autolens_workspace_test#253) — the literal alone would pass a
+source-plane mass regression. This block pins mass sensitivity directly; the floor
+is the audit-measured response divided by five (margin for platform drift).
+"""
+mass_indices = [
+    i
+    for i, name in enumerate(model.model_component_and_parameter_names)
+    if ".mass." in name
+    and "centre" not in name
+    and "ell_comps" not in name
+    and "redshift" not in name
+]
+assert mass_indices, "interferometer/datacube/rectangular: no mass parameters found for sensitivity check"
+
+parameters_perturbed = np.array(model.physical_values_from_prior_medians)
+for i in mass_indices:
+    parameters_perturbed[i] *= 1.05
+
+ll_median = float(np.asarray(result).ravel()[0])
+ll_perturbed = float(
+    np.asarray(fitness._vmap(jnp.array(parameters_perturbed[None, :]))).ravel()[0]
+)
+assert abs(ll_perturbed - ll_median) > 0.03, (
+    f"interferometer/datacube/rectangular: likelihood insensitive to a +5% lens-mass perturbation "
+    f"(median={ll_median}, perturbed={ll_perturbed}) — source-plane mass pipeline regression?"
+)
+print("PASS: mass-sensitivity floor exceeded.")
+
+
 """
 __Path A: jit-wrap parameter-vector entry point__
 

scripts/interferometer/jax_likelihood/delaunay.py

@@ -243,6 +243,40 @@ class in **PyAutoFit**, which pairs the model with likelihood.
     err_msg="interferometer/delaunay: JAX vmap likelihood mismatch",
 )
 
+
+"""
+__Mass Sensitivity__
+
+The literal above is evaluated at the model's prior medians, where a +5% change of
+every lens mass parameter moves this likelihood by less than the literal's rtol
+(audit 2026-08-06, autolens_workspace_test#253) — the literal alone would pass a
+source-plane mass regression. This block pins mass sensitivity directly; the floor
+is the audit-measured response divided by five (margin for platform drift).
+"""
+mass_indices = [
+    i
+    for i, name in enumerate(model.model_component_and_parameter_names)
+    if ".mass." in name
+    and "centre" not in name
+    and "ell_comps" not in name
+    and "redshift" not in name
+]
+assert mass_indices, "interferometer/delaunay: no mass parameters found for sensitivity check"
+
+parameters_perturbed = np.array(model.physical_values_from_prior_medians)
+for i in mass_indices:
+    parameters_perturbed[i] *= 1.05
+
+ll_median = float(np.asarray(result).ravel()[0])
+ll_perturbed = float(
+    np.asarray(fitness._vmap(jnp.array(parameters_perturbed[None, :]))).ravel()[0]
+)
+assert abs(ll_perturbed - ll_median) > 0.03, (
+    f"interferometer/delaunay: likelihood insensitive to a +5% lens-mass perturbation "
+    f"(median={ll_median}, perturbed={ll_perturbed}) — source-plane mass pipeline regression?"
+)
+print("PASS: mass-sensitivity floor exceeded.")
+
 # The sparse inversion path must obey the same lane-isolation contract as
 # imaging. Fitness sees the raw NaN after the forward pass and converts only
 # that lane to its configured resample value.

scripts/interferometer/jax_likelihood/delaunay_mge.py

@@ -252,6 +252,40 @@ class in **PyAutoFit**, which pairs the model with likelihood.
 )
 
 
+"""
+__Mass Sensitivity__
+
+The literal above is evaluated at the model's prior medians, where a +5% change of
+every lens mass parameter moves this likelihood by less than the literal's rtol
+(audit 2026-08-06, autolens_workspace_test#253) — the literal alone would pass a
+source-plane mass regression. This block pins mass sensitivity directly; the floor
+is the audit-measured response divided by five (margin for platform drift).
+"""
+mass_indices = [
+    i
+    for i, name in enumerate(model.model_component_and_parameter_names)
+    if ".mass." in name
+    and "centre" not in name
+    and "ell_comps" not in name
+    and "redshift" not in name
+]
+assert mass_indices, "interferometer/delaunay_mge: no mass parameters found for sensitivity check"
+
+parameters_perturbed = np.array(model.physical_values_from_prior_medians)
+for i in mass_indices:
+    parameters_perturbed[i] *= 1.05
+
+ll_median = float(np.asarray(result).ravel()[0])
+ll_perturbed = float(
+    np.asarray(fitness._vmap(jnp.array(parameters_perturbed[None, :]))).ravel()[0]
+)
+assert abs(ll_perturbed - ll_median) > 0.007, (
+    f"interferometer/delaunay_mge: likelihood insensitive to a +5% lens-mass perturbation "
+    f"(median={ll_median}, perturbed={ll_perturbed}) — source-plane mass pipeline regression?"
+)
+print("PASS: mass-sensitivity floor exceeded.")
+
+
 """
 __Path A: jit-wrap ``analysis.fit_from``__
 """

scripts/interferometer/jax_likelihood/mge_group.py

@@ -133,6 +133,40 @@
     err_msg="interferometer/mge_group: JAX vmap likelihood mismatch",
 )
 
+
+"""
+__Mass Sensitivity__
+
+The literal above is evaluated at the model's prior medians, where a +5% change of
+every lens mass parameter moves this likelihood by less than the literal's rtol
+(audit 2026-08-06, autolens_workspace_test#253) — the literal alone would pass a
+source-plane mass regression. This block pins mass sensitivity directly; the floor
+is the audit-measured response divided by five (margin for platform drift).
+"""
+mass_indices = [
+    i
+    for i, name in enumerate(model.model_component_and_parameter_names)
+    if ".mass." in name
+    and "centre" not in name
+    and "ell_comps" not in name
+    and "redshift" not in name
+]
+assert mass_indices, "interferometer/mge_group: no mass parameters found for sensitivity check"
+
+parameters_perturbed = np.array(model.physical_values_from_prior_medians)
+for i in mass_indices:
+    parameters_perturbed[i] *= 1.05
+
+ll_median = float(np.asarray(result).ravel()[0])
+ll_perturbed = float(
+    np.asarray(fitness._vmap(jnp.array(parameters_perturbed[None, :]))).ravel()[0]
+)
+assert abs(ll_perturbed - ll_median) > 0.005, (
+    f"interferometer/mge_group: likelihood insensitive to a +5% lens-mass perturbation "
+    f"(median={ll_median}, perturbed={ll_perturbed}) — source-plane mass pipeline regression?"
+)
+print("PASS: mass-sensitivity floor exceeded.")
+
 print("interferometer/mge_group.py checks passed.")
 
 

scripts/interferometer/jax_likelihood/rectangular.py

@@ -277,6 +277,40 @@
 )
 
 
+"""
+__Mass Sensitivity__
+
+The literal above is evaluated at the model's prior medians, where a +5% change of
+every lens mass parameter moves this likelihood by less than the literal's rtol
+(audit 2026-08-06, autolens_workspace_test#253) — the literal alone would pass a
+source-plane mass regression. This block pins mass sensitivity directly; the floor
+is the audit-measured response divided by five (margin for platform drift).
+"""
+mass_indices = [
+    i
+    for i, name in enumerate(model.model_component_and_parameter_names)
+    if ".mass." in name
+    and "centre" not in name
+    and "ell_comps" not in name
+    and "redshift" not in name
+]
+assert mass_indices, "interferometer/rectangular: no mass parameters found for sensitivity check"
+
+parameters_perturbed = np.array(model.physical_values_from_prior_medians)
+for i in mass_indices:
+    parameters_perturbed[i] *= 1.05
+
+ll_median = float(np.asarray(result).ravel()[0])
+ll_perturbed = float(
+    np.asarray(fitness._vmap(jnp.array(parameters_perturbed[None, :]))).ravel()[0]
+)
+assert abs(ll_perturbed - ll_median) > 0.008, (
+    f"interferometer/rectangular: likelihood insensitive to a +5% lens-mass perturbation "
+    f"(median={ll_median}, perturbed={ll_perturbed}) — source-plane mass pipeline regression?"
+)
+print("PASS: mass-sensitivity floor exceeded.")
+
+
 """
 __Path A: jit-wrap ``analysis.fit_from``__
 """

scripts/interferometer/jax_likelihood/rectangular_dspl.py

@@ -236,6 +236,40 @@ class in **PyAutoFit**, which pairs the model with likelihood.
 )
 
 
+"""
+__Mass Sensitivity__
+
+The literal above is evaluated at the model's prior medians, where a +5% change of
+every lens mass parameter moves this likelihood by less than the literal's rtol
+(audit 2026-08-06, autolens_workspace_test#253) — the literal alone would pass a
+source-plane mass regression. This block pins mass sensitivity directly; the floor
+is the audit-measured response divided by five (margin for platform drift).
+"""
+mass_indices = [
+    i
+    for i, name in enumerate(model.model_component_and_parameter_names)
+    if ".mass." in name
+    and "centre" not in name
+    and "ell_comps" not in name
+    and "redshift" not in name
+]
+assert mass_indices, "interferometer/rectangular_dspl: no mass parameters found for sensitivity check"
+
+parameters_perturbed = np.array(model.physical_values_from_prior_medians)
+for i in mass_indices:
+    parameters_perturbed[i] *= 1.05
+
+ll_median = float(np.asarray(result).ravel()[0])
+ll_perturbed = float(
+    np.asarray(fitness._vmap(jnp.array(parameters_perturbed[None, :]))).ravel()[0]
+)
+assert abs(ll_perturbed - ll_median) > 0.009, (
+    f"interferometer/rectangular_dspl: likelihood insensitive to a +5% lens-mass perturbation "
+    f"(median={ll_median}, perturbed={ll_perturbed}) — source-plane mass pipeline regression?"
+)
+print("PASS: mass-sensitivity floor exceeded.")
+
+
 """
 __Path A: jit-wrap ``analysis.fit_from``__
 """

scripts/interferometer/jax_likelihood/rectangular_mge.py

@@ -236,6 +236,40 @@ class in **PyAutoFit**, which pairs the model with likelihood.
 )
 
 
+"""
+__Mass Sensitivity__
+
+The literal above is evaluated at the model's prior medians, where a +5% change of
+every lens mass parameter moves this likelihood by less than the literal's rtol
+(audit 2026-08-06, autolens_workspace_test#253) — the literal alone would pass a
+source-plane mass regression. This block pins mass sensitivity directly; the floor
+is the audit-measured response divided by five (margin for platform drift).
+"""
+mass_indices = [
+    i
+    for i, name in enumerate(model.model_component_and_parameter_names)
+    if ".mass." in name
+    and "centre" not in name
+    and "ell_comps" not in name
+    and "redshift" not in name
+]
+assert mass_indices, "interferometer/rectangular_mge: no mass parameters found for sensitivity check"
+
+parameters_perturbed = np.array(model.physical_values_from_prior_medians)
+for i in mass_indices:
+    parameters_perturbed[i] *= 1.05
+
+ll_median = float(np.asarray(result).ravel()[0])
+ll_perturbed = float(
+    np.asarray(fitness._vmap(jnp.array(parameters_perturbed[None, :]))).ravel()[0]
+)
+assert abs(ll_perturbed - ll_median) > 0.02, (
+    f"interferometer/rectangular_mge: likelihood insensitive to a +5% lens-mass perturbation "
+    f"(median={ll_median}, perturbed={ll_perturbed}) — source-plane mass pipeline regression?"
+)
+print("PASS: mass-sensitivity floor exceeded.")
+
+
 """
 __Path A: jit-wrap ``analysis.fit_from``__
 """