Automatic reference audit · Program hashes, exact commands, outcomes and debugger evidence
The strict catalogue test remains FAIL. No emulator timing, acceptance image or pixel assertion was changed.
REU/reutiming2/c.prg draws short REU transfers into $d020. Its $d7ff=0 write marks a screenshot checkpoint, not a self-checked transfer result. Both original program revisions differ from the unchanged upstream PNG at exactly 337 pixels: normalized y=235, x=0–336 (raster 251), expected white (1), actual red (2).
The full original guest and component traces agree: the final byte $12 reaches $d020 at internal cycle 21 (PAL cycle 12), with BA high, no badline, $d011=$1b and one byte remaining. The call path is REU.DMAEvent.event → PLA.dmaWrite → VIC.write. All 32 existing final-write raster controls pass; the four catalogue controls produce three passes and this one failure.
Bauer’s VIC-II reference bounds ordinary badlines to rasters $30–$f7. The upstream image retains white for a badline-length interval on raster $fb, but its capture hardware is unidentified. Its GIMP metadata identifies an editor, not hardware. These observations do not establish a physical REU exception or prove the PNG wrong.
Fresh VICE 3.7.1 and 3.10 captures agree and also show red on the disputed strip. Their other REU/VIC arbitration differences remain substantial (6066 pixels versus JSIDPlay2), so VICE agreement does not establish hardware accuracy. Each version was reproduced twice with active SID emulation. These are software captures, not physical recordings.
The two optional historical-revision audits remain genuine failures in this evidence bundle and are separate from the default suite count. Resolving the discrepancy needs the unchanged program captured on a documented PAL C64/REU setup, ideally with BA and the last $d020 write recorded around raster $fb.






