Golden data regenerated: 2026-09-19, for 7.0.0~m12
=====================================================

Every file here was made by ctsim 6.0.2, built from the Debian source
package (ctsim_6.0.2-7) with its four patches applied and nothing else,
configured --without-x --disable-openmp so that only the text tools build.
It reports itself as "Version 6.0.0" because configure.ac in that release
was never updated; the tarball is 6.0.2.

The sizes are odd, because from 7.0.0~m12 the tools refuse even ones: 255
for 256, 127 for 128, 511 for 512, --nsample 3 for 2 and 5 for 4.  Even
golden files were removed rather than kept, there being nothing left that
could compare against them.

  herman-phm-255.if        phm2if 255 255 --nsample 3 --phantom herman
  herman-sino-367x320.if   phm2pj 367 320 --nray 2 --phantom herman; pj2if
  herman-rec-255.if        pjrec herman.pj 255 255 --filter abs_bandlimit
                           --filter-param 1
  herman-sinoEA-127x127.if phm2pj 127 127 --nray 1 --phantom herman
                           --geometry equiangular; pj2if
  sheplogan-phm-127.if     phm2if 127 127 --nsample 1 --phantom shepp-logan
  dicom-rec-255.if         pjrec dicom.pj 255 255 --filter abs_bandlimit
                           --filter-param 1

dicom-rec-255 is the one exception to "made by 6.0.2 alone": 6.0.2 has no
dicomimport, so the projections were imported by the current tree with
--legacy-format and the reconstruction -- the thing the file checks -- was
done by 6.0.2's pjrec.

physics.ref holds d, r and e of herman-phm-255 against herman-rec-255, in
that order, since the normalised distances are not symmetric.  6.0.2's own
if2 --comp gives 0.0562946 0.0221278 0.0523864; the current tree gives the
same at every digit 6.0.2 prints, and the file holds its fuller value.

Against these files the current tree agrees to 5e-7 or better on every
image, and bit-for-bit on both sinograms.  The tolerance is 1e-4.

pjfourier is deliberately not among the golden cases: 6.0.2 mishandled an
odd image there (its DC fell half a bin between two pixels), so its output
at these sizes is not a reference.

tests/data
----------
herman-255.png is herman-phm-255.if exported as 16-bit PNG with --min 0
--max 0.7, so that the raster phantom's scale of 0.7 maps the full range.
herman-511-ref.if is 6.0.2's phm2if 511 511 --nsample 5, and
herman-511.png is that exported the same way.  The window matters: exported
to the image's own range the raster phantom comes out a sixth too dense.
