1. Problem: Identifiers Bound to Storage, Not to Structure

Conventional content delivery and asset indexing systems reference digital assets by static identifiers: uniform resource locators, cryptographic hash pointers, or file-system paths derived from storage location or transmission metadata rather than from the internal structure of the content. Such identifiers are invalidated by mutation, format conversion, resolution change, lossy compression, or replication. The result is identity fragmentation across versions and derivatives, with no mechanism for derivative attribution, remix lineage, or variance-based similarity across stored objects.

The available repairs each fall short on their own terms. Perceptual hashing systems, including difference hash, average hash, and perceptual hash algorithms, produce low-dimensional binary signatures from downsampled representations; they lack multi-scale structural analysis, directional orientation decomposition, spatial sub-region identity, and a continuously scaled similarity score, and their fixed-width binary outputs cannot encode the gradient structure and compositional geometry needed for distributed anchor assignment. Watermarking and metadata tagging embed identity signals into the content stream or a sidecar record; watermarks are removable through transcoding, cropping, or generative reconstruction, and metadata records are decoupled from content structure and require persistent external storage. Blockchain-based asset registration anchors hash-based proofs to a ledger, inheriting the byte-equality problem, requiring global consensus, and binding identity to key-pair ownership rather than content structure.

The disclosure addresses this by computing a persistent, structurally derived, and policy-governed identity for digital content artifacts. Each artifact is assigned a unique identifier, or UID, derived deterministically from the artifact's internal variance and structural features rather than from its storage location, file name, cryptographic key, or transmission metadata. A UID encodes a multi-dimensional position within a continuous variance space, enabling similarity detection, mutation tracing, derivative attribution, and distributed governance without reliance on centralized registries, static addressing, or persistent credential infrastructure.

2. The Multi-Axis Variance Vector

The core construct is the multi-axis variance vector, extracted from the artifact's internal structure. The extraction pipeline accepts a digital content artifact and produces a nine-dimensional variance vector organized into three structural axes, designated X, Y, and Z. For raster images, the input is first converted to a normalized grayscale floating-point representation using a perceptual luminance weighting of approximately 0.299 red, 0.587 green, and 0.114 blue, yielding intensity values in the range zero to one. A multi-scale variance flow analyzer then subdivides the normalized scalar field into three nested grid resolutions: a coarse grid of 8x8 cells, a medium grid of 16x16 cells, and a fine grid of 32x32 cells. For each cell at each resolution the analyzer computes the variance of pixel intensity, then aggregates per-scale mean variance and standard deviation of variance.

The X-axis energy behavior vector encodes the artifact's energy distribution trend across scale, with three components: the slope of mean variance from coarse to fine resolution, the curvature of mean variance at the medium scale, and the asymptotic fine-scale energy value. The Y-axis frequency compaction vector encodes how variance dispersion behaves across scales, with three components: the rate of change of variance standard deviation from coarse to fine, the spread factor between maximum and minimum per-scale standard deviation, and the variance floor convergence value measuring proximity of scale-level mean variance to the artifact's global variance. The Z-axis structural phase persistence vector encodes orientation and structural stability, derived from a gradient histogram computed across eight angular bins spanning zero to pi radians, canonicalized so the dominant bin sits at index zero; its components are the horizontal-vertical orientation bias, the diagonal-axial bias, and a stability coefficient computed as one minus the absolute difference between edge density and normalized global variance.

This representation is designed to remain stable under format conversion, resolution rescaling within a defined canonical size, and lossy compression at moderate quality, while varying predictably under content-altering transformations such as object insertion, removal, significant cropping, style transfer, or compositional remixing. For non-image artifacts, modality-appropriate preprocessing produces a comparable normalized two-dimensional scalar field: audio is represented as a normalized mel-spectrogram, text as a token frequency scalar field, video as per-frame fields plus a temporal delta vector, and binary objects as a reshaped byte-variance field. All modalities converge on the same extraction stage.

3. UID Construction: Quadrants, Triads, and the 320-Bit Identifier

The nine-dimensional global vector is combined with quadrant sub-region vectors and optionally with structure and constellation signatures to produce the final UID. For hashing, a 27-dimensional vector is formed as a weighted linear combination of the X, Y, and Z axes organized into three dominant triads: an X-dominant triad weighting X at 0.6 and Y and Z each at 0.2; a Y-dominant triad weighting Y at 0.5, X at 0.3, and Z at 0.2; and a Z-dominant triad weighting Z at 0.5, X at 0.3, and Y at 0.2. This weighting ensures no single axis fully dominates the fingerprint and that it stays sensitive to perturbations along each behavioral dimension.

Spatial decomposition adds regional identity. Following canonical normalization to a 256-by-256 square canvas and an orientation canonicalization step that rotates the artifact when its dominant gradient orientation exceeds approximately 0.1 radians, the canonical image is divided into four non-overlapping quadrants. Each quadrant is run through the full nine-dimensional extraction and hashed into a 256-bit value using a coarsened quantization scheme, with X and Y components quantized at a step of 1/32 and Z components at 1/8 to absorb compression and format-conversion artifacts at the sub-image level. The four quadrant hashes are sorted in lexicographic order to canonical positions q0 through q3, which makes quadrant assignment independent of artifact orientation.

A multi-segment FNV-64 hash combiner applies five distinct initialization seeds to the ordered concatenation of the global 256-bit hash and the four sorted quadrant hashes, producing five 64-bit segments concatenated into a 320-bit UID. The first 16 hexadecimal characters, representing 64 bits, serve as a backward-compatible short-form identifier for display, indexing, and bandwidth-constrained transmission; the full 320-bit form is used for anchor assignment, slope-band routing, and similarity comparison. Two optional supplementary fingerprints extend matching: a gradient-only structure signature, invariant to mean luminance and background fill, that aids recognition of logos and graphically sparse artifacts; and a constellation signature, derived from the relative positions and angular relationships of high-salience anchor points, up to five detected over a 12x12 coarse saliency grid with at least three required for emission, and invariant to translation, rotation, and uniform scale change.

4. Slope Bands and Anchor Node Distribution

Identity must be resolvable at scale without a central directory. The variance band classification module assigns each UID to one of five variance bands based on the artifact's global variance value. Band 1 covers global variance below 0.02, characterizing near-uniform content such as solid fills and blank documents. Band 2 covers 0.02 to 0.06, low-detail content with sparse but identifiable structure. Band 3 covers 0.06 to 0.12, moderate-complexity content with visible regular structure. Band 4 covers 0.12 to 0.22, high-complexity content with varied spatial structure. Band 5 covers global variance at or above 0.22, very-high-variance content such as natural photographs and densely textured compositions.

Each band defines a routable index segment within the global slope continuum. Anchor nodes are assigned to one or more bands and declare governance responsibility for UIDs whose global variance falls within their band scope. An anchor node is not a fixed network endpoint; it is a governance entity that derives its routing responsibility from its declared band scope. A band adjacency graph records which bands share governance boundaries, enabling cross-band resolution for content that drifts between bands under mutation. Because any processing node that knows a target UID's variance value can determine which band governs it, resolution queries route to the appropriate anchor cluster without consulting a central registry, which keeps resolution functional in disconnected, asynchronous, and adversarial environments. Alternative deployments may define finer band granularity, such as 10, 20, or 100 bands, or a continuous slope spectrum with fuzzy boundaries, without modifying the UID structure.

Within each band, anchor nodes form a quorum-based resolution mesh. Each anchor stores a fragment of the UID index for its bands and participates in an Adaptive Consensus Protocol that validates alias registrations, resolves conflicting UID assignments, and maintains routing consistency. The protocol uses trust-weighted asynchronous voting in which vote weight derives from declared band scope, historical reliability, and trust-zone authority; per-mutation quorum thresholds configurable by mutation type, so routine registration requires a configurable minimum quorum while structural mutations such as anchor recruitment or band split require a configurable supermajority; and lineage-preserving commitment in which each accepted mutation records its predecessor state, participating anchor signatures, and the policy version under which it was evaluated.

5. Cosine Similarity and Multi-Root Lineage Graphs

A single UID locates an artifact in variance space; a graph of UIDs connected by similarity edges records how artifacts relate. The platform supports composite content attribution by allowing each UID to participate in a directed lineage graph that may contain more than one parent. When a derivative artifact is registered against one or more candidate parents, the variance vector comparison module computes the cosine similarity between the derivative's variance vector and each parent's vector, defined as the inner product divided by the product of the magnitudes, computed over the concatenated X, Y, and Z components of the nine-dimensional global vector. A slope delta computation module computes the Euclidean distance between the candidate and each proposed parent, normalized over a configured maximum distance to yield a similarity score in the range zero to one.

These scores determine whether the observed slope proximity falls within the configured semantic continuity threshold required to establish a lineage edge. Confirmed parents receive a contribution weight proportional to their cosine similarity with the derivative, recorded as an edge annotation. The disclosure is explicit that these weights are structural signals that may inform licensing, attribution display, and policy inheritance; they do not constitute legal determinations of authorship or ownership. Versioning is structural rather than declarative: any new UID falling within the configured slope continuity threshold of a previously registered UID is automatically a candidate for version linkage, subject to anchor quorum confirmation.

The comparison framework produces per-axis cosine similarity scores for X, Y, and Z, an aggregate directional cosine similarity, a Euclidean distance-based score, and per-quadrant similarity scores with hash match flags. Quadrant-level comparison supports localized mutation detection: a derivative that modifies only one spatial region of a source exhibits quadrant similarity near 1.0 for unchanged regions and reduced scores for the modified region, enabling spatially resolved attribution even when global similarity remains high. Adversarial recombinations, in which a party attempts to suppress attribution to a source by manipulating variance features, are detectable because the composite UID's slope profile diverges measurably from the weighted variance combination of its declared parents; anchors may flag such entries for heightened review.

6. Structural Lineage Absence and Synthetic Content

The disclosure exploits a structural property that distinguishes synthetically generated content from captured content: a generatively synthesized artifact has no structural lineage to any prior registered artifact in the anchor network. When an image, audio clip, or video sequence is produced by a diffusion model, a generative adversarial network, or a language model over visual tokens, the resulting artifact's position in slope space reflects the statistical properties of the generative model's output distribution rather than the variance profile of any specific prior artifact.

The lineage query module detects this by querying the anchor network for registered parent UIDs within a configured slope continuity radius of the candidate's UID. If no registered parent falls within the radius, the orphan detector classifies the artifact as structurally unanchored: it has no provable lineage connection to any content registered in the governed corpus. The disclosure is careful here. Structurally unanchored artifacts are not necessarily fraudulent or impermissible, but they cannot be admitted under a policy object that requires verifiable provenance, and they trigger heightened scrutiny under policy objects that govern synthetic content.

Two further structural detectors operate from the candidate's own features. The screenshot recapture classifier evaluates the Z-axis horizontal-vertical orientation bias against a policy-calibrated threshold; screen rendering and recapture introduce periodic spatial-frequency structure that manifests as elevated energy in the horizontal and vertical orientation bins relative to the diagonal bins, producing a recapture probability score without any reference to the original artifact. The synthetic content detector compares the candidate's variance vector against a slope-band-indexed statistical model of known synthetic content distributions and produces a synthesis probability score. A composite risk score aggregator combines lineage absence, recapture probability, and synthesis probability into a governance signal routed to the pre-release admissibility engine.

7. Rights-Grade Admissibility at the Commitment Boundary

The same variance-derived identity supports governance of generative and distributable content at the commitment boundary rather than through post-hoc moderation. A commitment is any irreversible or externally visible side effect of a generation or distribution event, including public release, customer delivery, API return, licensing, marketplace publication, training data admission, or cross-platform provenance anchoring. The pre-release admissibility engine interposes an evaluation between the generation of a candidate artifact and its commitment, so that structurally impermissible content cannot become a released artifact.

A candidate artifact is routed through two parallel tracks before a commitment gate. The policy object evaluator tests the artifact against one or more versioned, cryptographically signed, machine-evaluable policy objects defining typed category constraints, jurisdictional scopes, override authorities, similarity tolerance thresholds, and escalation paths. The structural similarity evaluator computes cosine similarity between the candidate's variance vector and the vectors of reference artifacts indexed in a governed corpus, which is itself a slope-band-indexed anchor network registered under signed corpus policy objects. If similarity to any exclusion-corpus entry exceeds the policy-declared threshold, the candidate is rejected, regenerated under modified constraints, or escalated to an override authority. Because similarity operates over variance-derived UIDs rather than GPU inference or centralized embedding indexes, it can run client-side, at generation time, without per-query compute proportional to corpus size.

Admissibility decisions are reproducible and auditable: given the variance-derived UID and structural signatures of the evaluated artifact and the policy object version, any authorized party may replay the evaluation and compare against the recorded decision, in distinction from opaque classifiers. A training corpus governance layer admits artifacts only under signed corpus policy objects and records a governance record for each, comprising the variance-derived UID, the governing policy object, a timestamp, and a cryptographic hash of the policy object, constituting a verifiable lineage from trained model artifacts back to the admissible corpus. A consultation event logger deterministically records, for each generation event that consults a reference artifact through retrieval-augmented generation or structured neighborhood resolution, the consulted artifact's UID, the governing policy object, the variance proximity score, and a timestamp.

8. Client-Side Execution and Substrate Deployment

The architecture is substrate-agnostic and operates independently of device-bound session state, transport-layer addressing, and centralized certificate authorities. UIDs may be computed on any conforming node, stored in memory-native anchor substrates, and resolved through variance-band-routed queries without external registry consultation, supporting deployment across centralized data centers, federated institutional clusters, decentralized peer networks, mobile edge nodes, and intermittently connected devices.

In a client-side embodiment, a file object or media stream is received through a standard browser file input or media capture API, normalized using only the Canvas 2D API for canonical resizing, grayscale conversion, and orientation canonicalization, with no WebGL, WebAssembly, or server-side dependency. The variance vector computation, 27-dimensional vector construction, and 320-bit UID hashing run using standard JavaScript arithmetic and the FNV-variant hash combiner. Local similarity evaluation runs against a slope-band-filtered exclusion corpus fragment pre-fetched from the anchor network, and a locally stored signed policy object; both are verifiable by their cryptographic signatures without a live connection at evaluation time. Only the computed UID and the resulting admissibility decision leave the device; the raw content artifact does not, which conforms to data-minimization requirements and avoids per-query inference costs.

Resolution itself is exposed through a stateless, policy-scoped query protocol. A querying client submits a candidate UID, its policy scope identifier, and a timestamp, and receives a resolution response containing policy-permitted matches. The protocol defines identity resolution for exact or near-exact matches, derivative resolution for matches between the continuity and identity thresholds, orphan resolution for an empty match set, and conflict resolution that routes overlapping lineage claims to the fork adjudication handler. Bulk resolution lets a client submit batches of UIDs across multiple bands in a single request, routed in parallel to their band clusters. The disclosure characterizes "near real time" as producing a result with a slight but acceptable delay, in the range of about 250 milliseconds.

9. Disclosure Scope

The computable content identity platform, comprising the multi-axis variance vector extraction pipeline over nested 8x8, 16x16, and 32x32 grids; the X, Y, and Z structural axes; the 27-dimensional axis-dominant triad weighting; quadrant decomposition with rotation-invariant lexicographic sorting; the multi-segment FNV-64 combiner producing a 320-bit UID with a 64-bit short form; the optional structure and constellation signatures; the five variance bands and their global-variance boundaries; anchor node distribution under the Adaptive Consensus Protocol; cosine-similarity multi-root lineage graphs with weighted parent attribution; orphan detection, screenshot recapture detection, and synthetic content detection; the rights-grade pre-release admissibility engine, training corpus governance layer, and consultation event logger; and the client-side execution and stateless resolution query protocol, is disclosed in PCT International Application No. PCT/US26/28630. This article describes that disclosed mechanism.

The disclosure does not assert universal robustness against all transformations. Operations that exceed the artifact's stability envelope, including significant cropping, generative substitution, and compositing of unrelated artifacts, produce a divergent variance vector and may migrate the UID to a different slope band; this is expected behavior, and continuity is then maintained through the lineage graph rather than through UID invariance. Lineage contribution weights are structural signals and do not constitute legal determinations of authorship, ownership, or infringement. Memorization proximity and similarity scores are structural measurements calibrated against policy-declared thresholds, not adjudications. The modality-specific normalization recipes are illustrative: any normalization producing a bounded two-dimensional scalar field of normalized values may serve as input to the extraction pipeline without modification to the downstream hashing, slope-band assignment, anchor governance, or lineage construction components.