[us-commits] [ehb54/ultrascan3] ed1512: extrap_c0: warn when fit_broaden is ignored; keep ...

[emre brookes]

  Branch: refs/heads/ehb54-issue-920
  Home:   https://github.com/ehb54/ultrascan3
  Commit: ed15126d5c2b3a05f5f03cee6a3e404857003123
      https://github.com/ehb54/ultrascan3/commit/ed15126d5c2b3a05f5f03cee6a3e404857003123
  Author: ehb54 <brookes at uthscsa.edu>
  Date:   2026-07-23 (Thu, 23 Jul 2026)

  Changed paths:
    M us_somo/develop/src/us_hydrodyn_saxs_iqq_extrap_c0.cpp

  Log Message:
  -----------
  extrap_c0: warn when fit_broaden is ignored; keep it out of the name

The manual fit-broadening window applies only to the additive model with automatic GCV
off and no reference-scale/splice. In every other case it was silently accepted and
silently dropped -- most importantly it is INERT in reciprocal mode (the default), so a
user could set it, see no effect, and have no idea why. The output name also gained a
_fbN token even when the window did nothing, falsely advertising a broadened run.

Now: when fit_broaden > 1 is requested but will not be used, log exactly why (reciprocal
model / GCV on / reference-scale or splice on). And the name is built from
fit_broaden_effective -- the value that actually took effect -- so _fbN appears only when
the window was really applied.

Found while sweeping regularization strength as a mechanism test: fit_broaden turned out
to change nothing in reciprocal, which is what exposed the silent no-op.

Verified on all four cases: reciprocal / additive+GCV / additive+ref-scale each warn and
drop the _fb token; plain additive+nogcv applies it and keeps _fb9. All 9 regression
baselines byte-identical (none use fit_broaden).

Co-Authored-By: Claude Fable 5 <noreply at anthropic.com>



To unsubscribe from these emails, change your notification settings at https://github.com/ehb54/ultrascan3/settings/notifications