|

Glossary

Short definitions for the terms that turn up across the docs and the specification.


Roles and emitters

Source role
The source_role on an event says who observed it. origin is the content owner's web server (e.g. a WordPress plugin), edge is a CDN or edge network (Cloudflare, Fastly), index is a content marketplace or licensed repository, and agent is the AI agent itself. Several observers can report the same retrieval from different vantage points. See the API reference.
Org roles - content owner / platform / agent
The registry has three kinds of organisation. A content owner publishes content and reads telemetry about it (self-serve onboarding). A platform is anything that handles content or telemetry on others' behalf - marketplaces, networks, vendors, analytics tools (gated onboarding). An agent is an AI system that retrieves and cites content (gated onboarding). See Platforms and Agent integration.
Delegation
A content owner granting a platform read access to its telemetry. The platform queries the same content-owner endpoints with delegated_from={grantor_org_id}. Only the grantor can revoke. See Measurement partners.
Cross-observer correlation
Joining events that several emitters reported about the same retrieval - say an edge worker and an agent both seeing the same page fetch. The shared content_telemetry_id is what links them when there is no session context.

Levels

Conformance level - retrieval / grounding / attribution
How much of the model an emitter implements. Retrieval: report content_retrieved only, no session needed. Grounding: also report which content was loaded into the agent's context. Attribution: full session - retrieval through citation, engagement, and outcome. It is informational and emitter-side; it describes what an emitter does, not a gate it has to pass.
Privacy level - full / summary / intent / minimal
How much detail an emitter includes about a turn or session. full: the prompt and response text. summary: a description, not the raw text. intent: just the intent category. minimal: nothing beyond which content was involved. Emitters pick the level they are comfortable with.

Content stages

A piece of content moves through up to five stages in a session, each its own event type:

Retrieved (content_retrieved)
The agent (or an edge / origin / index observer) fetched the content. The one stage you can see without agent cooperation.
Grounded (content_grounded)
The content was loaded into the agent's context to inform the response.
Cited (content_cited)
The content was referenced in the agent's response. This is the anchor for outcome attribution.
Displayed (content_displayed)
The content was shown to the user in a card, sidebar, or similar surface.
Engaged (content_engaged)
The user interacted with the cited content - clicked through, expanded it, and so on.

Identifiers and headers

content_telemetry_id / Content-Telemetry-ID
A correlation id minted per retrieval. The agent can send it as the Content-Telemetry-ID request header when it fetches a page so an edge or origin emitter can stamp the same id on its event - which lets the two observations be joined later. It carries no user identity.
ctx token
A short-lived token (ctx_..., 90 days) an agent creates for an outbound link. Appended as ?ctx=... on the destination URL, it lets the landing site tie the click back to the session - which content was retrieved and cited. Lookup at GET /ctx/{token} is two-sided opt-in. See the API reference.
content_url vs content_id
content_url is the canonical URL of the content - the usual identifier on events. content_id is an optional stable identifier for content that has one independent of a URL (a DOI, an ISBN), used mostly by marketplace and index emitters.

The manifest

Manifest / .well-known/openattribution.json
A small JSON file a participant publishes at /.well-known/openattribution.json declaring who they are, their role, contact and policy links, and optional signing keys. For content owners it doubles as one of the domain-verification methods. Format is in section 8 of the specification; see also The .well-known file.
manifest_ref
A field on a session pointing at the URL of a manifest (root or path-prefixed). It links the session to the publishing participant - useful when content lives under a path with its own manifest.