Expensive to get wrongSouth KoreaUnited StatesBeauty & personal care

Software can buy from the US shelf and not the Korean one

41 of 47 storefronts selling into the United States publish a machine-readable commerce manifest at “/.well-known/ucp” declaring which operations an autonomous shopping agent may run against them, 41 of those declaring a checkout capability, while 0 of 19 storefronts read in the Korean home market publish one at all.

A shopping agent asked to buy a Korean product can complete the purchase on a US-facing storefront and cannot complete it on the brand’s home one, so the brand’s own market is the market its products are hardest to buy from by machine, and the gap widens on its own every time the US platform ships a capability.

Limits how to read Korean storefronts publish no product data, US ones do

74 of a possible 90Tier B, systematic first-party observation. Nothing at this tier is allowed past 90, however good it looks.

Verified September 9, 2026Rechecked every 60 days, next by November 8, 2026.

Structure

Held since August 5, 2026How the market is physically or commercially arranged. The earliest date we are willing to say this held. It may well have been true before that; we do not backdate a record past the evidence we have read. Supported, version 5.

This entry is wrong if

A repeat run finds any Korean home-market storefront in this set publishing an agent commerce manifest, or finds fewer than half the US-facing set publishing one.

Written before the record was published, not after. If you can show this, send it to hello@tryorbt.com and the correction goes on the record with the entry.

Evidence

Tier Bobserved-webn = 66
orbt: Market Register commerce collector, run commerce-20260805140548
Read at the path RFC 8615 reserves for capability discovery, on every platform in the set, because a 404 here is the storefront’s answer rather than a failed read. One thing this does not say: on the US side the manifest is the platform’s and not the merchant’s, and storefronts run by unrelated companies returned byte-identical documents, so the adoption number is a fact about the platform those brands chose. The corridor gap is the finding, not any individual brand’s decision. Declared protocol version 2026-04-08; COSRX declares 8 capabilities. Nothing was added to a cart and no advertised agent endpoint was called. Run commerce-20260805140548.
Read 2026-08-05 · CC BY · page hash 9a6ccfd5ea85

History

v5correctedAugust 5, 2026
Correction. The retirement recorded above was an operator error, not a finding. The derivation was run against collector run commerce-20260805053004 while that crawl was still in flight and had reached 4 of 66 storefronts, so every claim whose storefronts it had not yet read failed its sample gate and was retired. 44 claims were retired in one command for that reason. The claim was re-derived the same day from collector run commerce-20260805051110, a complete 66 of 66 crawl, and holds. Nothing about the market changed between the two records. The retirement stays in this log because the log is not edited; this entry is what it means. The instrument now refuses to derive from a run missing more than a tenth of the seeded storefronts, and a claim returning from retired now files its own status revision, which is why this correction had to be written by hand rather than appearing on its own.
v4retiredAugust 5, 2026
Retired against collector run commerce-20260805053004: sample too thin (US 2, KR home 0). The claim held on the sample available when it was written and does not hold on this one.
What it said before
41 of 47 storefronts selling into the United States publish a machine-readable commerce manifest at “/.well-known/ucp” declaring which operations an autonomous shopping agent may run against them, 41 of those declaring a checkout capability, while 0 of 19 storefronts read in the Korean home market publish one at all.
v3correctedAugust 5, 2026
Re-derived from collector run commerce-20260805051110 after a change to the instrument: added two page families to the instrument: the blog sitemap child plus Atom feed, which measures what a storefront actually publishes rather than whether the platform gave it a blog, and robots.txt read as data, which records which crawlers a shop lets in and reads identically on Cafe24 and Shopify The figures moved for that reason. Any part of the change attributable to the market moving cannot be separated from it in this revision, and the previous statement is recorded below rather than replaced.
What it said before
41 of 47 storefronts selling into the United States publish a machine-readable commerce manifest at “/.well-known/ucp” declaring which operations an autonomous shopping agent may run against them, 41 of those declaring a checkout capability, while 0 of 18 storefronts read in the Korean home market publish one at all.
v2correctedAugust 5, 2026
Re-derived from collector run commerce-20260805040124 after a change to the instrument: replace markdown backticks with quotation marks in claim prose; the site and every downstream consumer render claim text as plain prose, so a backtick was showing up literally in a sentence meant to be quotable The figures moved for that reason. Any part of the change attributable to the market moving cannot be separated from it in this revision, and the previous statement is recorded below rather than replaced.
What it said before
41 of 47 storefronts selling into the United States publish a machine-readable commerce manifest at `/.well-known/ucp` declaring which operations an autonomous shopping agent may run against them, 41 of those declaring a checkout capability, while 0 of 18 storefronts read in the Korean home market publish one at all.
v1createdAugust 5, 2026
Derived from collector run commerce-20260805040124.

Append-only. Nothing in this list is edited or removed once written, including the entries that record our own mistakes. Every change to every record is on the corrections page, and the same log is in entries.json under revisions.

agent-commerce
ucp
protocol
platform
machine-readability
ai-visibility
corridor
k-beauty
beauty

Cite this entry

“Software can buy from the US shelf and not the Korean one”, The Orbt Market Index 0.1.0, entry omr:kr-us:only-one-side-of-the-corridor-can-be-bought-from-by-software, retrieved September 9, 2026, https://tryorbt.com/markets/entry/kr-us--only-one-side-of-the-corridor-can-be-bought-from-by-software. Dataset: https://doi.org/10.5281/zenodo.22679129

Orbt, "Software can buy from the US shelf and not the Korean one," Orbt Market Index, entry omr:kr-us:only-one-side-of-the-corridor-can-be-bought-from-by-software, verified September 9, 2026, https://tryorbt.com/markets/entry/kr-us--only-one-side-of-the-corridor-can-be-bought-from-by-software.

@misc{orbt-kr-us-only-one-side-of-the-corridor-can-be-bought-from-by-software,
  title        = {Software can buy from the US shelf and not the Korean one},
  author       = {{Orbt}},
  year         = {2026},
  doi          = {10.5281/zenodo.22679129},
  note         = {The Orbt Market Index 0.1.0 (dataset DOI 10.5281/zenodo.22679129), entry omr:kr-us:only-one-side-of-the-corridor-can-be-bought-from-by-software. Verified 2026-09-09},
  howpublished = {\url{https://tryorbt.com/markets/entry/kr-us--only-one-side-of-the-corridor-can-be-bought-from-by-software}},
  license      = {Orbt Index Public Licence. Free to quote and cite; not to redistribute}
}

For a deck, a brief or a chat.

Free to quote, cite and act on. Not to resell or redistribute — the sources ask that of us, and each one’s terms are named. Machine-readable: entries.json carries this record under the id omr:kr-us:only-one-side-of-the-corridor-can-be-bought-from-by-software, with the same evidence and falsifier.

This entry is wrong if

Evidence

History

Cite this entry

Read next