From b28305dd250b3382bc39a94806e6c85ec4de49f5 Mon Sep 17 00:00:00 2001 From: Barely Removable Date: Sat, 22 Aug 2026 09:59:47 -0700 Subject: [PATCH] Grading: distinguish miscut/trimming from ordinary corner wear; fix surface over-hedging and over-calling --- vision.py | 56 +++++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 56 insertions(+) diff --git a/vision.py b/vision.py index 42bd6ef..497f11b 100644 --- a/vision.py +++ b/vision.py @@ -278,6 +278,45 @@ drive the category — graders do not average corners. white paper showing through a coloured border, or a visibly blunted or \ bent tip — not a uniformly soft edge. If a close-up is simply too blurry \ to tell the difference, say cannot_assess. + A corner that looks out-of-square — not rounded or frayed, but genuinely \ +cut or angled differently from the other three — needs a different read than \ +ordinary wear, and three distinct things can produce it: + PHOTO ANGLE, rule this out first. A shot taken even slightly off-axis \ +compresses and skews the far corners relative to the near ones, exactly \ +mimicking a bad cut. If the card's outline looks like a parallelogram in a \ +way that follows the perspective of the shot — corners nearer the camera \ +reading "wider" — that is the photo, not the card. Do not report it as a \ +defect. + A genuine MISCUT is a factory cutting error: the border runs visibly \ +wider on one side than centering alone would explain, part of an adjacent \ +card's image bleeds in at one edge, or a border is entirely missing on one \ +side. PSA's MC qualifier is REQUIRED whenever this is present — it is never \ +optional and a miscut card never gets a plain numeric grade, only one with \ +MC attached. Say so explicitly when you see this pattern (it has to affect \ +the design-to-edge relationship generally, not just look like one corner is \ +off), since a miscut can coexist with an otherwise pristine card and \ +qualifier cards are conventionally valued around two grade-points below an \ +unqualified card at the same number — worth naming even though it isn't \ +reflected in the numeric estimate itself. + TRIMMING is more serious than either of the above and much rarer: it is \ +post-manufacture alteration, done to fake sharper corners or edges than the \ +card actually has, and PSA refuses ANY numeric grade once it's identified — \ +not a low grade, no grade at all. The tells are specific, so only raise this \ +when you actually see them, not from a single corner looking slightly odd: \ +corners that appear to curl or point UPWARD off the card's plane (PSA's own \ +term for this is a "bat-ear" cut), edges or corners that are glossy, \ +unnaturally sharp, or hooked compared to the card's own factory cut \ +elsewhere, or a cut that looks wavy/rippled rather than flat. A single \ +photo cannot confirm trimming the way calipers and UV light can, so treat it \ +as a possibility worth flagging in the note for the owner to check by hand \ +— phrase it as "worth checking for trimming" rather than a verdict, but do \ +name the specific visual tell that prompted it. + Absent any of the above, a single corner that's mildly not-square while \ +the rest of the card reads as a clean rectangle is ordinary manufacturing cut \ +tolerance — common enough, especially pre-1980s — and not worth flagging on \ +its own. Reserve this whole paragraph for when you can actually see one of \ +the three specific patterns above, not as a reason to hedge on an otherwise \ +normal-looking corner. - EDGES: look for whitening, nicks, chipping, or roughness along the four \ edges. Assess top, right, bottom and left separately and let the worst one \ drive the category — graders do not average edges. @@ -317,6 +356,23 @@ loss of gloss, or foil/holo scratching. reverse-holo cards; and blocky square patterns from image compression. If \ the right panel is uniformly busy rather than showing distinct lines, that \ is print texture, not damage. + That uniformly-busy, texture-not-damage case is a "none" finding for that \ +quadrant — you looked, you could tell it apart from real damage, so say so \ +plainly. That is a different thing from cannot_assess, which is for when the \ +PHOTO fails you (real glare washing out a region, motion blur, a quadrant too \ +dark to make out either panel) rather than for when the panels are readable \ +but nothing damage-like stands out. Treat every SURFACE INSPECTION pair \ +you're given as sufficient, by default, to reach a real severity for that \ +quadrant — hold cannot_assess to the same bar you'd apply to a corner or edge \ +close-up that is genuinely too blurry to read, not to one that's merely busy. + Corroborate before calling something damage, the same direction as the \ +crease-vs-print-line rule below: a mark that shows up in the right (processed) \ +panel but that you cannot also point to, even faintly, in the left \ +(untouched) panel is more often a compression artefact or noise from the \ +processing than a real scratch. The right panel finds candidates; the left \ +panel confirms them. Don't report a right-panel-only mark as more than \ +"minor", and don't report it at all if you can't describe where it sits in \ +the untouched crop too. PRINT LINES are the important exception to that, and they are easy to \ dismiss as holo texture when they are not. A print line is a STRAIGHT band \ running parallel to one edge, usually spanning most or all of the card's \