[krystophny]

direct_fl_int projects each field-line point back onto its target flux surface, but previously stored the uncorrected ODE radius in y_out. The resulting rzphi chart disagreed with the corrected coordinates used for field evaluation. Store the projected radius in the chart while preserving the ODE state and integration tolerances.

Validation:

• The focused regression runs the existing circular EQDSK through read_eq_efit and checks the delivered chart against the input flux function. Its maximum normalized flux residual is 2.45438e-5 before the fix and 2.90926e-6 after it (threshold 1e-5).
• With the same TC24 EQDSK, mpsi=512, mtheta=1024 and etol=1e-8, reconstructed Bpol error against the input field falls from 3.43968e-4 to 1.88200e-5 in L2, and from 3.59914e-3 to 4.25633e-4 in maximum norm. Toroidal flux is unchanged; q and the consumer's |B| remain stable. The remaining field error still exceeds the benchmark targets and is tracked separately.
• Native reader build and focused regression pass; independent review found no scope or test issue. git diff --check passes.

Chris&AI

[krystophny]

The complete projected-radius correction is integrated into the canonical GPEC production fork through PR6, squash 91808b3. It also projects the current upstream common-angle samples and passes the existing delivered-chart oracle at its unchanged tolerance. The maintained upstream report355 and this original source branch remain retained.

Chris&AI