How Audiveris and oemer were run

Both were run in one container image, in batch, over the staged page images,
one process per page, default settings, no per-page tuning, no manual
correction, first attempt only. The exact invocations:

    Audiveris 5.11.0
        <audiveris> -batch -export -output <out_dir> <image>
      MusicXML is taken from the exported .mxl (the first non-META-INF entry),
      falling back to any .xml containing <score-partwise>.

    oemer 0.1.x
        oemer <image> -o <out_dir> --without-deskew

    Timeouts: 600 s for Audiveris, 2400 s for oemer. A page that timed out,
    crashed or produced no parseable MusicXML is recorded as no output and
    scores zero rather than being dropped.

Known limitation, and the reason this file exists: the *photographed* suite was
staged to this runner as page one of each piece only. Audiveris and oemer were
therefore never shown the second page of the eighteen two-page pieces, and the
first published photograph figures scored them against ground truth including
those pages. All published figures now score page one only, for every engine.
score.py --all-pages reproduces the earlier multi-page numbers.

Run dates: Audiveris and oemer 2026-07-15; Sheetly 2026-08-22. Both competitor
versions are the current releases of those projects and unchanged since.
