Mechanism
The platform supports composite content attribution by enabling each unique identifier to participate in a directed lineage graph that may contain more than one parent. A content artifact generated through combination, transformation, remixing, or derivation from multiple prior artifacts is registered under a new unique identifier linked to all of its contributing source unique identifiers, with each lineage edge weight reflecting the degree of variance inheritance from that contributor. The graph is multi-root because the derivative is linked to each declared source whose variance structure is measurably proximate to it, not to a single ancestor.
When content artifact C is registered as a derivative of content artifact A and content artifact B, a variance vector comparison module computes the cosine similarity between the variance vector of C and the variance vectors of A and B respectively. Cosine similarity is computed over the concatenated X, Y, and Z axis components of the nine-dimensional global variance vector as the inner product of the two vectors divided by the product of their magnitudes. The same variance vectors that anchor each artifact's identity are the vectors compared here, so no separate representation is constructed for the purpose of attribution.
A slope delta computation module computes the Euclidean distance between the variance vector of the candidate derivative and each proposed parent, normalized over a configured maximum distance threshold to produce a similarity score in the range zero to one. These similarity scores determine whether the observed slope proximity is within the configured semantic continuity threshold required to establish a lineage edge. A proposed parent whose proximity is within the threshold is linked; a proposed parent whose proximity falls outside it is not.
Weighted Parent Attribution
A weighted parent attribution stage assigns a contribution weight to each confirmed parent unique identifier proportional to its cosine similarity with the derivative unique identifier. These weights are recorded as edge annotations in the directed lineage graph. They do not constitute legal determinations of authorship or ownership. They serve as structural signals that may inform licensing, attribution display, policy inheritance, and governance enforcement by downstream systems. A directed lineage edge registration module commits the edges, the slope delta values, the contribution weights, and the anchor endorsement signatures to the lineage graph stored in the governing anchor's cache memory.
Because the weights are derived from the measured cosine similarity between the derivative and each parent rather than from a publisher's assertion of contribution, the attribution is a property of the recorded variance relationship rather than of a declaration a publisher is free to set. A downstream system that allocates value or enforces consent in proportion to contribution can read the edge weights and enumerate the parent set directly from the committed edges.
Structural Versioning
A version history record stores a directed graph of unique identifier transitions ordered by registration timestamp. Each edge encodes the mutation type, the slope delta between the predecessor and successor unique identifiers, and the policy inheritance conditions applicable to the mutation. Versioning under this model is structural and does not depend on author declaration or explicit version numbering. Any new unique identifier that falls within the configured slope continuity threshold of a previously registered unique identifier is automatically considered a candidate for version linkage, subject to anchor quorum confirmation.
A delegation policy evaluator enforces ownership delegation constraints at the time of unique identifier registration and mutation. Delegation policy objects specify permitted mutation types, temporal validity bounds, unique identifier lineage depth limits, trust zone whitelists, and transferability conditions. If a proposed mutation violates a delegation constraint, for example where an unauthorized agent attempts to register a derivative unique identifier from a policy-protected source, a mutation rejection handler denies registration and records the attempt in the anchor event log. Delegation constraints may propagate through version lineages in accordance with the inheritance mode specified in the policy object.
Spatially Resolved Attribution
The comparison framework operates on unique identifier records and produces per-axis cosine similarity scores for the X, Y, and Z axes, an aggregate directional cosine similarity, a Euclidean distance-based similarity score, and per-quadrant similarity scores together with hash match flags. The per-quadrant comparison supports localized mutation detection. A derivative artifact that modifies only one spatial region of a source exhibits quadrant similarity scores near one for the unchanged regions and reduced scores for the modified region, so attribution can be resolved to a spatial region even when the global similarity score remains high.
Quadrant-level resolution distinguishes this attribution from a single global match. A composite that draws one region from one source and another region from a different source does not present as a near-duplicate of either source taken whole, yet the per-quadrant scores expose which region tracks which parent. The attribution is therefore not merely a list of similar artifacts but a structurally grounded account of where each contribution appears.
Detection of Adversarial Recombination
A composite content artifact is considered fully auditable. Any party with access to the governing anchor cluster may trace the unique identifier of a composite artifact to its contributing parents, inspect mutation metadata, evaluate policy lineage, verify anchor quorum endorsements, and assess whether the composition satisfies the governance requirements of the applicable trust zones.
Adversarial recombinations, in which a party attempts to register a derivative that suppresses attribution to one or more source unique identifiers by manipulating variance features, are detectable because the slope profile of the composite unique identifier will exhibit measurable divergence from the weighted variance combination of its declared parents. Construction of the lineage graph includes verifying that the variance vector of a declared composite artifact is consistent with the weighted variance combination of its declared parent artifacts within a configured tolerance, and flagging as potentially adversarial any composite artifact whose variance profile exhibits discontinuity from that weighted combination beyond the configured tolerance. Anchors may flag such entries and subject them to heightened governance review.
Composition with the Anchoring Layer
Composite lineage builds directly on the content anchoring layer that assigns each artifact a variance-derived unique identifier. The provenance validator constructs and traverses multi-root lineage graphs by computing slope vector proximity and mutation deltas between a candidate unique identifier and one or more previously anchored unique identifiers, the same anchored identifiers the platform already maintains. The lineage graph fragments, alias registrations, and mutation records reside in the cache memory associated with each anchor node, governed within that node's variance band scope.
Because lineage edges are committed with anchor endorsement signatures under anchor quorum consensus, the attribution is preserved and verifiable without reliance on a centralized registry, static network addressing, or persistent external credentials. The lineage construction, like the unique identifier derivation and slope band assignment beneath it, is a function of the content's measured structure and the governing anchors rather than of an external authority.
Prior-Art Distinctions
Watermarking and metadata tagging bind provenance to an embedded marker or a sidecar record. Watermarks are removable through transcoding, cropping, or generative reconstruction, and metadata records are decoupled from content structure. The disclosed mechanism computes the lineage relationship from the content's own variance structure, so a source that did not embed a marker is still detectable as a contributor, and a composite cannot launder the provenance of its sources by stripping a marker.
Perceptual hashing systems produce a fixed binary signature for near-duplicate detection. The variance vector here is structured for cosine comparison and slope delta computation across axes and quadrants, which supports weighted multi-parent attribution and spatially resolved contribution rather than a single duplicate flag. Because each parent edge is populated from the measured cosine similarity and slope delta between the composite's variance vector and a candidate predecessor's variance vector, the lineage edges are derived from structure rather than asserted by a publisher.
Disclosure Scope
The composite attribution and multi-root lineage model, comprising the registration of a derivative unique identifier linked to multiple contributing source unique identifiers, the computation of cosine similarity and normalized slope delta between the derivative's variance vector and each proposed parent's variance vector, the establishment of lineage edges where proximity is within the configured semantic continuity threshold, the weighting of each edge proportional to cosine similarity, the structural version history record, the delegation policy enforcement at registration and mutation, the per-quadrant and per-axis comparison framework, and the detection of adversarial recombination through divergence from the weighted variance combination of declared parents, is disclosed in PCT International Application No. PCT/US26/28630. This article describes that disclosed mechanism.
The scope extends to embodiments that vary the modality of the content, the band granularity of the anchor network, and the inheritance mode of delegation policy, provided that lineage edges are populated from measured variance proximity between a candidate unique identifier and previously anchored unique identifiers and that the composite's consistency with the weighted combination of its declared parents is verified. Specific licensing, attribution-display, and rights-management applications built atop the lineage graph are outside the scope except as exemplars, and the edge weights are structural signals rather than legal determinations of authorship or ownership.
