Graded scan pins wrong catalog identity: EN Fossil Articuno 2/62 PSA slab resolved to Japanese ipb_ja-2 (スリープ/Drowzee) via graded_cache number collision #538

New Issue

2026-08-30T15:38:01Z

gambit-admin commented

2026-08-30 15:38:01 +00:00

Prod scan at 2026-08-30 15:14 UTC (tester IP 12.50.143.136). PSA-slabbed 1999 Fossil Articuno-Holo 1st Edition 2/62, PSA 9, cert 79165854.

The graded lane itself worked: cert read from the slab QR (cert_src=qr), Card Hedge fmv-by-cert resolved the cert to '1999 Pokemon Fossil Articuno-Holo 1st Edition 2' (card=True fmv=True), match=high, custody-check calls show the app rendered the slab.

THE BUG — wrong catalog identity pinned:
app.routes.external_scan: graded-scan identity veto (graded_cache) label='articuno'/'2/62' dropped=[('bp-2','Hitmonchan'),('pl3-2','Blaziken FB'),…]
app.routes.external_scan: graded-scan ocr-first identity game=pokemon card_id=ipb_ja-2 via=graded_cache number='2' name='スリープ'

ipb_ja-2 is a JAPANESE card named スリープ (Drowzee), a bare number-2 collision — not Fossil Articuno. The client then received card_id=ipb_ja-2 (its own follow-up: GET /external/scrydex-listings?...&name=Articuno&card_id=ipb_ja-2&company=PSA&grade=9), so the PSA-9 comps ran against the wrong card, and detail/image/identity downstream is the Japanese Drowzee wearing the cert's 'Articuno' display title. To the vendor it looks like the scan matched some random plain card.

Mechanism (backend/app/routes/external_scan.py ~L3970-4060): EN Fossil Articuno 2/62 apparently isn't in the sparse graded price cache, so after the label veto + language gate the only surviving number-hit was the JP row, and the 'one distinct card = unambiguous' acceptance pinned it. The code comment already warns this gate is weak ('a bare number often has exactly one cached holder and looks unambiguous while being flatly wrong'). Two gates failed to save it: _graded_label_contradicts didn't treat スリープ as contradicting 'articuno' (non-Latin name comparison?), and _graded_language_filter didn't drop a Japanese card on an English label read.

Suggested direction: (a) on an English-script label read, a JP-set candidate should never satisfy the unambiguous single-holder gate without name alignment; (b) treat a non-Latin cache name vs a Latin label name as a contradiction in the veto; (c) when the cache has no name-aligned survivor, fall through to Scrydex search rather than cache-pinning a number collision. Note the Card Hedge cert title ('…Fossil Articuno-Holo 1st Edition 2') was in hand and names the set — it could anchor/veto the identity choice.

Filed from a Claude Code session

Prod scan at 2026-08-30 15:14 UTC (tester IP 12.50.143.136). PSA-slabbed 1999 Fossil Articuno-Holo 1st Edition 2/62, PSA 9, cert 79165854. The graded lane itself worked: cert read from the slab QR (cert_src=qr), Card Hedge fmv-by-cert resolved the cert to '1999 Pokemon Fossil Articuno-Holo 1st Edition 2' (card=True fmv=True), match=high, custody-check calls show the app rendered the slab. THE BUG — wrong catalog identity pinned: app.routes.external_scan: graded-scan identity veto (graded_cache) label='articuno'/'2/62' dropped=[('bp-2','Hitmonchan'),('pl3-2','Blaziken FB'),…] app.routes.external_scan: graded-scan ocr-first identity game=pokemon card_id=ipb_ja-2 via=graded_cache number='2' name='スリープ' ipb_ja-2 is a JAPANESE card named スリープ (Drowzee), a bare number-2 collision — not Fossil Articuno. The client then received card_id=ipb_ja-2 (its own follow-up: GET /external/scrydex-listings?...&name=Articuno&card_id=ipb_ja-2&company=PSA&grade=9), so the PSA-9 comps ran against the wrong card, and detail/image/identity downstream is the Japanese Drowzee wearing the cert's 'Articuno' display title. To the vendor it looks like the scan matched some random plain card. Mechanism (backend/app/routes/external_scan.py ~L3970-4060): EN Fossil Articuno 2/62 apparently isn't in the sparse graded price cache, so after the label veto + language gate the only surviving number-hit was the JP row, and the 'one distinct card = unambiguous' acceptance pinned it. The code comment already warns this gate is weak ('a bare number often has exactly one cached holder and looks unambiguous while being flatly wrong'). Two gates failed to save it: _graded_label_contradicts didn't treat スリープ as contradicting 'articuno' (non-Latin name comparison?), and _graded_language_filter didn't drop a Japanese card on an English label read. Suggested direction: (a) on an English-script label read, a JP-set candidate should never satisfy the unambiguous single-holder gate without name alignment; (b) treat a non-Latin cache name vs a Latin label name as a contradiction in the veto; (c) when the cache has no name-aligned survivor, fall through to Scrydex search rather than cache-pinning a number collision. Note the Card Hedge cert title ('…Fossil Articuno-Holo 1st Edition 2') was in hand and names the set — it could anchor/veto the identity choice. --- Filed from a Claude Code session

gambit-admin added the bug api labels 2026-08-30 15:38:01 +00:00

gambit-admin added the claimed:kris label 2026-08-30 15:55:36 +00:00

gambit-admin closed this issue

2026-08-30 17:05:59 +00:00

gambit-admin removed the claimed:kris label 2026-08-30 17:05:59 +00:00

gambit-admin referenced this issue

2026-09-02 21:01:00 +00:00

JP graded slabs (SV2a) fail to resolve despite card + Master Ball variant existing in card_cache #578

gambit-admin referenced this issue

2026-09-02 23:09:40 +00:00

card_cache has no language column — EN/JP prints share game='pokemon' and every name+number reader is language-blind #580

gambit-admin referenced this issue