Rendered from docs/decisions/0015-a-synthesized-content-tier.md in the Headwater corpus. Every document on this half of the site is typed by the taxonomy the descriptor names: corpus.json.

Q15 — A synthesized content tier

Context

One evaluation settles this with Q19 and Q18. All three ask what a provenance record must carry for content that the corpus did not author, and what such content may then do.

The open-question entry named the wrong field, and its own text says so without following through. It says that "the boundary that this question draws is a verification method, not an author". It then argues about the author for the rest of its length. Agency was never the boundary. Spec 3 already admits an agent-drafted document in full: agency: agent, drafted_by, and a human in accepted_by. What the entry actually names is content that nobody accepted, at a volume where acceptance does not scale.

Decision

Every document carries a warrant: the mechanism by which the corpus can defend that the document is what it claims to be (spec 1). The set is closed and it has four values.

Warrant	What stands behind the content
accepted	A named human accepted it
regenerated	It is a function of inputs inside the repository, and generate --check proves it
transcribed	It is a byte-faithful copy of a pinned external snapshot (Q19)
asserted	Nothing

Yes, Headwater admits asserted content. The mark is positive and never an absent field, which is what SPDX settled when it made NOASSERTION a value beside NONE. Wikipedia is the observed case at the largest available scale. Its 2025 speedy-deletion criterion for machine output fires on the absence of review, and not on the presence of a model (HW-EVAL-adjacent-work §P).

Consequences

What it may not do is derived, and the leaning's list is replaced. Two rules cover the cases that "never a valid target for governs or verifies" enumerated, plus every relation that an adopter adds later.

• No edge may let unwarranted content govern the reading of warranted content. Reading precedence is already derived from nuclearity, succession and the governance family, so no per-relation list is needed. Anchors carry no reading precedence, so an asserted document may still declare governs against a code path. That edge is what gives it write-time impact detection, which is the only thing that will report its decay.
• An asserted document does not discharge an evidence obligation. evidenced obliges an external, auditable artifact, and an asserted document is neither (spec 3).

Staleness does not apply, and there is no date to bump. last_verified records that a human confirmed a document. An asserted document has none and cannot acquire one without becoming accepted. Its decay is reported as drift from its production date, and the remedy is to regenerate or to delete.

Promotion is acceptance, and nothing new is built. A person reads the document, sets the warrant, and names themselves. Bulk stamping produces identical bytes, so no mechanism detects it. That posture is advisory permanently, for the reason that the shift ratio is.

The instrument is sited differently from the way this record proposed, and the build is what found that. This record said that taxonomy audit reports promotions per change. That verb reads one working tree, and a promotion is a lifecycle transition whose count needs the prior version of each changed document. Spec 12 makes that input available only in change-scoped evaluation, and it names needs_prior as the declaration that asks for it. warrant.promoted declares it, and headwater check --change is where a caller supplies the change. So the reading belongs to a change-scoped check rather than to this verb, and it is built there. What taxonomy audit reports is the standing asserted population, which is the denominator and not the rate. Spec 13 carries what that leaves the decision owing.

The Q17 constraint changed the answer rather than confirming it. The open-question entry proposed that an emitter which cannot carry the mark declares that in its loss set. A loss-set entry reaches the consumer who reads it and nobody else. C2PA's own threat model records that ordinary tooling strips a mark that travels beside content. So an emitter that cannot carry the warrant withholds the content at the profile's declared tombstone grain (spec 6). That costs nothing today, because the native export carries the warrant with no loss.