Position coord on CorrectedMonitor causes transmission_fraction to fail

[SimonHeybrock]

Summary

transmission_fraction in ess.sans.normalization fails with a DatasetError (coordinate mismatch on position) when computing the transmission fraction using monitors from the GenericTofWorkflow path (i.e., via LokiWorkflow).

The root cause: monitor_to_wavelength calls transform_coords('wavelength', graph=graph, keep_inputs=False), but position is not an input to the MonitorCoordTransformGraph. The graph starts from pre-computed Ltotal, so position is a bystander coordinate that survives the conversion. The transmission_fraction provider then multiplies ratios from incident and transmission monitors, which sit at different physical positions, causing scipp to reject the operation.

Reproduction

The error occurs when:

• Monitor data is loaded from a NeXus geometry file (via assemble_monitor_data), which attaches position from EmptyMonitor
• The data flows through monitor_to_wavelength → preprocess_monitor_data → CorrectedMonitor
• transmission_fraction computes (sample_transmission / direct_transmission) * (direct_incident / sample_incident) — each division preserves the respective monitor's position, and the final multiply fails

scipp._scipp.core.DatasetError: Mismatch in coordinate 'position' in operation 'multiply':
()    vector3              [m]  (0, 0, 0.2)
vs
()    vector3              [m]  (0, 0, -8.4)

Verification

import scipp as sc

# Simulates what monitor_to_wavelength does
da = sc.DataArray(
    sc.ones(dims=['tof'], shape=[10]),
    coords={
        'tof': sc.linspace('tof', 1000, 50000, 10, unit='us'),
        'position': sc.vector([0, 0, -8.4], unit='m'),
        'Ltotal': sc.scalar(25.0, unit='m'),
    }
)

from scippneutron.conversion import graph as conv_graph
graph = {
    **conv_graph.beamline.beamline(scatter=False),
    **conv_graph.tof.elastic_wavelength('tof'),
    'source_position': lambda: sc.vector([0, 0, -60.0], unit='m'),
}

result = da.transform_coords('wavelength', graph=graph, keep_inputs=False)
print(list(result.coords))  # ['position', 'wavelength'] — position survives!

Impact

This affects any workflow using GenericTofWorkflow (e.g., LokiWorkflow) that computes TransmissionFraction with monitors at different physical positions. The LokiAtLarmorWorkflow is unaffected because its custom larmor_monitor_to_tof conversion path doesn't attach position the same way.

Possible fixes

• Drop position explicitly in monitor_to_wavelength after the transform_coords call.
• Or include position in the MonitorCoordTransformGraph so it gets consumed (and thus dropped by keep_inputs=False).
• Or drop non-dimension coords in preprocess_monitor_data / transmission_fraction.

Workaround

We currently work around this in ESSlivedata by stripping non-wavelength coords when mapping CorrectedMonitor values between run types.