[tomneiser]

pf_active_coil_current_data in omas/machine_mappings/d3d.py reads the F1A..F9B PTDATA signals and divides current.data (and data_error_upper) by element.0.turns_with_sign (55–58), with the comment that IMAS stores the current per turn. IMAS does store the current per turn, but the DIII-D F-coil PTDATA signal is the power-supply (conductor) current, i.e. already per turn; the amp-turns are PTDATA × turns. After the division the ODS carries F-coil currents 55–58× too small.

Check. A free-boundary Grad-Shafranov refit of the EFIT02 boundary of shot 200000 at t = 3.04 s (FUSE/FRESCO, Green functions = per-turn current × turns_with_sign) converges to F6A = −227.7 kA-turns = −4.14 kA × 55, and likewise on every F-coil (F1A −138, F6B −257, F7B −181 kA-turns), i.e. the undivided PTDATA values (F6A −4.12 kA) within 1–3 %. Started from the divided values the same refit reaches the same solution but needs ~50× more iterations (85 s vs 1.6 s).

History: the division was added in 2cac7d5 (2023-08-29, first in v0.92.0, message "MDS+ stores A-turns"), but its condition tested the path string f'pf_active.coil.{channel}.identifier' (never contains an uppercase F), so it was dead code until 7625664 (2026-04-08, first in v0.96.0) changed it to ods1[...identifier]. ODSs from checkouts before that commit therefore carry the correct PTDATA values; everything exported with v0.96.0 or later carries F-coils 55–58× too small.

Change. The division is removed; the loop keeps only the missing-data warning. turns_with_sign stays as the element turns. E-coils (ECOILA/B, E567/E89 segments) have one turn per element and are not affected.

Where the "A-turns" came from. The EFIT k-file and the EFIT MDS+ tree do hold the F-coil currents in amp-turns for DIII-D (BRSP in OMFIT's samples/k133221.01000 is −53 to −142 kA-turns on F1A..F3A), and OMFIT's omfit_eqdsk.py (ff769052a0b, same day as 2cac7d5) converts those to the IMAS per-turn current by dividing by turns_with_sign on the way in and multiplying on the way out. That pair is correct for an ODS that holds the per-turn current. The PTDATA pointnames F1A..F9B that this mapping reads are the power-supply current, i.e. already per turn, so the division here produces an ODS 55–58× below what the OMFIT converters and the free-boundary solvers expect. A k-file written by OMFIT from an ODS exported with v0.96.0+ comes out with BRSP = PTDATA instead of PTDATA × turns, and update_keqdsk_from_omas.py (which plots ODS current × turns against the k-file) shows the same factor. With this fix the OMFIT side needs no change.

Downstream: FUSE PR ProjectTorreyPines/FUSE.jl#1195 adds a loader repair for ODSs exported with the divided currents.

🤖 Generated with Claude Code

[smithsp]

Makes sense to me

[smithsp]

Waiting for @torrinba to also approve

[torrinba]

I agree that this was a mistake. Thanks for catching and fixing it @tomneiser

I think the fix is actually needed for equilibrium.time_slice[:].constraints.pf_current, but since that isn't part of d3d.py it's not quite so simple to add. I agree that it shouldn't be present for pf_active.coil[:].current.data and probably mixed these up in the previous commits

I expect this fix is needed in imas_composer as well

[AreWeDreaming]

@torrinba the toksearch version introduces efit.py and gets rid of the _efit.json direct MDSplus mappings (if I recall correctly). So in principle any efit mapping should basically act like d3d.py from there onwards