The raster comparison engine (§6.3): locate the entviz, gate on shape/size,
then sample each cell's nucleus against the render model and disprove on a
mismatch. Disprove-only — different (a predicted nucleus color is wrong) or
unknown (couldn't analyze / no visible difference); never identical (we
can't OCR the text, and a nucleus color is a lossy function of the token).
reference is the decoded pasted image (RGBA); value/opts are ours. The
note is stripped from our model so a reference that differs only by a note still
compares equivalent — we anchor on the frame's width + top and sample only the
grid, so the bottom note/suffix strip is never read.
The raster comparison engine (§6.3): locate the entviz, gate on shape/size, then sample each cell's nucleus against the render model and disprove on a mismatch. Disprove-only —
different(a predicted nucleus color is wrong) orunknown(couldn't analyze / no visible difference); neveridentical(we can't OCR the text, and a nucleus color is a lossy function of the token).referenceis the decoded pasted image (RGBA);value/optsare ours. The note is stripped from our model so a reference that differs only by a note still compares equivalent — we anchor on the frame's width + top and sample only the grid, so the bottom note/suffix strip is never read.