fix: swap source/target grids in XsecRecord::Extract interpolation

src/core/absorption/xsec_fit.cc

@@ -161,7 +161,7 @@ void XsecRecord::Extract(VectorView result,
     {
       const auto f_gp =
           lagrange_interp::make_lags<1, lagrange_interp::grid_identity>(
-              f_grid_active, data_f_grid_active, 0.5, "Frequency");
+              data_f_grid_active, f_grid_active, 0.5, "Frequency");
       const auto f_itw = reinterpweights(f_gp);
 
       // Find frequency grid positions:

tests/python/xsec/xfit.py

@@ -1,30 +1,52 @@
-import pyarts3 as pyarts
 import numpy as np
+import pyarts3 as pyarts
+
+# x_ref: pinned strided values (stride 60000) of the native-grid
+# spectral_propmat output for O3 at STP. Generated once from the
+# native-grid case and frozen as the regression reference.
+x_ref = [
+    5.48830541e-29,
+    7.01126943e-26,
+    2.25566466e-25,
+    3.67157516e-27,
+    5.42679027e-26,
+    7.56822053e-23,
+    1.11608430e-21,
+    3.90151892e-22,
+]
+
+# Tolerances: 1e-36 is the native-grid round-trip floor (effectively
+# exact). The resampled grids go through np.linspace, which introduces
+# interpolation error, so we loosen to 1e-18 for those cases.
+cases = [
+    (1.0, 1e-36),  # native grid; exact
+    (0.2, 1e-18),  # downsampled; interpolation error dominates
+    (1.2, 1e-18),  # upsampled;   same note as downsampled
+]
+
+
+def _check_grid(ws, atm, nd, f, stride, atol):
+    x = ws.abs_xfit_data.spectral_propmat(f=f, atm=atm, spec="O3") / nd
+    assert np.allclose(x[::stride, 0], x_ref, atol=atol)
+
+
+def main():
+    ws = pyarts.Workspace()
+
+    atm = pyarts.arts.AtmPoint()
+    atm.pressure = 101325.0
+    atm.temperature = 273.15
+    atm["O3"] = 1e-8
+    nd = atm.number_density("O3-666")
+
+    ws.abs_speciesSet(species=["O3-XFIT"])
+    ws.ReadCatalogData()
+    f = ws.abs_xfit_data["O3"].fitcoeffs[0].grids[0]
+
+    for n, atol in cases:
+        f_n = f if n == 1.0 else np.linspace(f[0], f[-1], int(f.size * n))
+        _check_grid(ws, atm, nd, f_n, stride=int(60000 * n), atol=atol)
+
 
-ws = pyarts.Workspace()
-
-atm = pyarts.arts.AtmPoint()
-atm.pressure = 101325.0
-atm.temperature = 273.15
-atm["O3"] = 1e-8
-nd = atm.number_density("O3-666")
-
-ws.abs_speciesSet(species=["O3-XFIT"])
-ws.ReadCatalogData()
-f = ws.abs_xfit_data["O3"].fitcoeffs[0].grids[0]
-x = ws.abs_xfit_data.spectral_propmat(f=f, atm=atm, spec="O3") / nd
-
-assert np.allclose(
-    x[::60000, 0],
-    [
-        5.48830541e-29,
-        7.01126943e-26,
-        2.25566466e-25,
-        3.67157516e-27,
-        5.42679027e-26,
-        7.56822053e-23,
-        1.11608430e-21,
-        3.90151892e-22,
-    ],
-    atol=1e-36,
-)
+if __name__ == "__main__":
+    main()