Replies: 1 comment
|
Reading the current published specification literally, the answers are:
The sentence about filter-only browse requests being returned ‘without text-relevance ranking’ does not imply a neutral order. It only says that there was no text query from which to compute text relevance; the business still chooses the order. The catalog overview also says that ordered arrays put the most relevant item first and that, for search, the best match is based on query and context—without specifying or exposing the scoring function. Relevant current sections:
I cannot answer whether the working group plans this beyond the published roadmap, but I do not see evidence that it is intentionally guaranteed to remain opaque either. Today it is simply unspecified by the base capability. The protocol already has a clean way to add it: a negotiated Catalog Search extension. For example, an extension could add: {
"ranking": {
"requested_objective": "user_preference",
"preferences": ["camera_quality", "brand"]
}
}and a per-result disclosure such as: {
"placement": {
"kind": "sponsored",
"sponsor": "seller-or-campaign-id"
},
"ranking": {
"criterion": "business_relevance",
"organic_position": 4
}
}For this to solve the trust problem you describe, the extension would need a normative rule such as MUST disclose when payment or another commercial incentive changed inclusion or order. Making the fields optional would not let a buyer agent distinguish ‘organic’ from ‘undisclosed.’ The request side should also distinguish ‘apply these user preferences’ from ‘return a business-neutral order’; the latter may be impossible for some implementations, so the response needs an explicit supported/applied value rather than silent fallback. Until such an extension is negotiated, a buyer agent should treat result order as business-selected—not as evidence of neutral quality—and re-rank returned candidates itself using product attributes. That still cannot recover products displaced from the returned page, which is why disclosure/request semantics belong at the protocol level rather than only in agent policy. |
Uh oh!
There was an error while loading. Please reload this page.
I'm evaluating UCP for a buyer-side agent that shops on behalf of a user with an explicit ranking preference (e.g., "camera quality matters more than brand").
Before asking, I checked
docs/specification/catalog/search.mdand the search request/response scaffolds: the only ranking-related language I found is that filter-only queries return products "without text-relevance ranking," and I couldn't find any field marking a result as sponsored, boosted, or otherwise commercially incentivized. Apologies if I missed it in another schema.Three concrete questions:
Context for why I'm asking: an agent that picks the top-ranked result satisfying its hard constraints will systematically absorb the business's ranking preferences into what looks like a neutral choice. If ranking incentives aren't machine-readable, the agent can violate its user's stated preference order without any protocol violation on either side. I'd like to understand whether that is considered in scope for UCP's catalog capability, or intentionally left to marketplace regulation / implementer policy. Either answer helps.
All reactions