Mechanism

Rights-grade inference governance is a governance layer that evaluates candidate inference transitions for compliance with creator attribution requirements, content exclusion mandates, and intellectual property governance constraints before those transitions are committed. It is not a separate authorization service that sits in front of the inference engine. It operates inside the same inference-time semantic execution substrate that governs every other transition: each candidate inference transition is mapped to a structured mutation descriptor, and the rights-grade layer evaluates that descriptor as part of the admissibility determination rather than inspecting the finished output after generation.

The defining property is that rights are evaluated at the transition level rather than at the output level. A candidate transition that would introduce rights-encumbered content is evaluated at the moment it is proposed, before it is committed and before it conditions any subsequent transition. Because each committed transition conditions the transitions that follow it, evaluating rights at proposal time prevents rights-encumbered content from entering the inference trajectory at all, rather than detecting it once it has already shaped the rest of the output.

Placement Within the Admissibility Gate

The rights-grade governance layer operates within the policy constraint evaluation stage of the semantic admissibility gate. The admissibility gate evaluates each proposed mutation through sequential stages and produces one of three deterministic outcomes: admit, reject, or decompose. Policy constraint evaluation is the first stage because policy violations are absolute, and rights constraints are evaluated there alongside domain, safety, structural, and task-specific policies.

Because rights enforcement lives inside the gate, it inherits the gate's structural guarantees. The evaluation is deterministic: given the same semantic state object and the same proposed mutation, the determination is the same. The evaluation is recorded: every admissibility determination, including a rights-driven rejection and the constraint it violated, is written to the lineage field of the semantic state object. And the evaluation is co-located with the transition it governs, so there is no separate path by which a transition can advance without passing through the rights layer.

The Rights Governance Sub-Field

The constraints the layer enforces are carried in a rights governance sub-field of the policy reference field. The policy reference field already encodes the structured set of governance policies that apply to the inference operation: domain policies specifying authorized and excluded semantic domains, safety policies specifying content-level constraints, structural policies specifying format and scope requirements, and task-specific policies supplied by the invoking agent. The rights governance sub-field is the part of that structure that names rights obligations.

The rights governance sub-field specifies three kinds of constraint. It specifies the content domains for which creator attribution is required. It specifies content exclusions, which prohibit reproduction or substantial incorporation of specific identified works, styles, or patterns. And it specifies provenance requirements, which state the degree to which derivation from training data must be documentable. These three constraint kinds, attribution, exclusion, and provenance, are the substance of what the rights layer evaluates each transition against.

Attribution and Exclusion Evaluation

Evaluation begins when a candidate transition's mutation descriptor indicates content matching a rights-governed domain. The match is determined by semantic similarity evaluation: the layer compares the content the transition would introduce against the rights-governed domains named in the sub-field. A transition whose content does not match any rights-governed domain proceeds through the rest of the admissibility gate without rights evaluation, so the layer imposes its checks only where rights are actually implicated.

When a candidate transition does match a rights-governed domain, the layer evaluates it for both attribution compliance and exclusion compliance. A transition that fails exclusion compliance, meaning it would reproduce or substantially incorporate an excluded work, style, or pattern, is rejected. A transition that satisfies exclusion compliance but fails attribution compliance is handled according to the governing policy: it may be admitted with an attribution annotation, or it may be rejected. The choice between annotated admission and rejection is itself a policy decision carried in the rights governance sub-field, not a fixed behavior of the layer.

Provenance and the Training-Time Governance Loop

The provenance constraints in the rights governance sub-field connect inference-time rights evaluation to the governance that occurred when the model was trained. The platform records a training provenance log that captures, for each training example, the policy object that authorized its admission, the depth profile applied to it, and its content provenance record. The inference substrate can query this log to determine the training-time governance profile of the knowledge that grounds a candidate transition.

When the inference substrate identifies output similar to a known training artifact, it can consult a training-level memorization assessment that classifies the similarity as shallow memorization, deep memorization, or absent memorization. If the assessment indicates shallow memorization of properly governed content, the substrate may permit the output with an attribution annotation. If it indicates deep memorization of content that should have been depth-restricted, the substrate may suppress the output and generate a governance alert. If the training content that grounds a transition was admitted under a policy that has since expired, the substrate may apply heightened scrutiny or reject the transition; if the policy was revoked by the content owner, the substrate may reject the transition and generate a governance alert. The rights layer therefore reflects not only what the model is about to say but the governance under which the knowledge behind it was acquired.

Distinction from Post-Generation Rights Systems

Post-generation copyright and content-identification systems analyze a completed output for similarity to known works. They can flag or suppress an output after it exists, but they cannot prevent the inference engine from generating the similar content in the first place, and they cannot prevent that content from conditioning the transitions that follow it. By the time a post-generation system inspects the output, the rights-encumbered content has already shaped the remainder of the inference trajectory.

The disclosed layer evaluates each transition at proposal, preventing rights-encumbered content from entering the inference trajectory rather than detecting it afterward. This is the same structural advantage the semantic execution substrate provides generally, applied to rights: governance is interposed within the inference loop rather than wrapped around its output, and the determination, whether admit, reject, or annotated admission, is recorded in the lineage as a first-class governed event.

Composition with the Admissibility Gate

Because the rights layer sits inside the policy constraint evaluation stage, it composes naturally with the gate's other stages. A transition that survives rights evaluation still faces mutation descriptor validation, lineage continuity validation, and entropy bounds evaluation; a transition that fails rights evaluation is rejected at the first stage and never reaches the others. Rights compliance is thus a precondition expressed in the same vocabulary as every other admissibility criterion, evaluated by the same gate, against the same semantic state object.

The rights determination also participates in the lineage record that the substrate maintains for the whole inference process. Each lineage entry carries the admissibility determination and, for a rejection, the evaluation stage at which it occurred and the specific constraint violated. A rights rejection or an annotated admission is recorded with the same structure as any other determination, so an auditor reviewing the lineage sees not only the transitions that ran but the rights-governed transitions that were refused or annotated and the rights constraint that drove each outcome.

Disclosure Scope

The rights-grade governance layer, comprising the rights governance sub-field of the policy reference field that specifies creator attribution domains, content exclusion mandates, and provenance requirements; the evaluation of a candidate inference transition for attribution compliance and exclusion compliance when its mutation descriptor matches a rights-governed domain determined by semantic similarity; the rejection of a transition that fails exclusion compliance and the policy-governed admission with attribution annotation or rejection of a transition that fails attribution compliance; the transition-level rather than output-level operation of the layer; and the recording of each rights determination in the lineage field, is disclosed in the cognition filing (U.S. Application No. 19/647,395 and its international counterpart) at Section 8.17. This article describes that disclosed mechanism.

The same filing further discloses, in its training-governance chapter, embodiments in which the rights layer consults the training provenance log and the training-level memorization assessment to enrich its determination, including admission with attribution annotation for shallow memorization of properly governed content, suppression and governance alert for deep memorization of content that should have been depth-restricted, and rejection of transitions grounded in expired or revoked content policy. The defining property in every such embodiment is that rights compliance is evaluated at the transition level within the admissibility gate, before commitment, and that the determination is recorded as a first-class governed event in the lineage.