NF2 segment choice for evaluations

Use the same physical component family for training and evaluation unless the experiment is specifically testing a conversion path.

For normal HMI SHARP CEA NF2 work, use:

Br,Bp,Bt,Br_err,Bp_err,Bt_err

Training

Use Br, Bp, and Bt as the vector magnetic-field boundary. Include Br_err, Bp_err, and Bt_err when available so the workflow can account for measurement uncertainty.

Time-series runs

Use the same CEA component family. extrapolate_series.py searches for files with names like:

*Bp.fits
*Bt.fits
*Br.fits
*Bp_err.fits
*Bt_err.fits
*Br_err.fits

Boundary-quality evaluation

The NF2 bottom slice is in local model components:

Bx, By, Bz

The SHARP CEA boundary is in:

Bp, Bt, Br

So the comparison mapping is:

(Bx, By, Bz) <-> (Bp, -Bt, Br)

State this mapping explicitly in any boundary-agreement plot or metric.

What not to mix

Do not compare an NF2 bottom slice directly against field/inclination/azimuth. Those segments must first be converted into the same vector-component convention.

Similarly, do not use SST-style B_los/B_trv/B_azi files with the SHARP CEA loader. That is a coordinate-language mismatch, not a clever shortcut.

Defaults

For AR 11158 HMI work: use SHARP CEA selected by HARP number, with:

Br,Bp,Bt,Br_err,Bp_err,Bt_err