Try any of 2778 endpoints — live.

Search 2nd STREET. ONE ROW IS ONE PHYSICAL SECOND-HAND PIECE held by ONE NAMED SHOP — the chain buys outright and resells, so there is no seller, no variant and no stock count; the size IS the row and every Japanese row carries the `shop_id` that holds it. JAPAN: 60 rows a page, fixed (the source swallowed per_page/limit/count and served 60 when 120 was asked for), and the WHOLE result set is walkable — page 20,000 of the unfiltered catalogue still returned 60 real rows and the page after the last one returns an EMPTY grid, never a repeat. Prices are whole yen and TAX-INCLUDED: `税込` appears on every product page measured and `税抜` on none, and the grid price was checked against 12 product pages' own microdata (12/12 identical). Japanese text — size, condition, colour, material — is republished exactly as the source wrote it; nothing is translated. UNITED STATES: the storefront's search answers at most 1,000 results (it prints '1000 results found' for nike, jacket and even `*`) over 28 pages of 36, so a large query ends with `stop_reason: cap_reached`. Pass `collection` instead of `keyword` to walk a collection without that cap, up to 250 rows a page; that surface additionally publishes `tags` (which carry the US branch name), `images` and `compare_at_price_usd`, which the keyword surface does not print.