Δx, Δy
Placement
Signed offset of a structural-field centroid from frame center.
Not automatically the depicted subject, gaze fixation, or a judgment against centering.
The instrument
Coordinates make a visual relationship negotiable.
VTL does not decide that centered is bad, complexity is good, or stillness is empty. It measures a rendered surface so a person—or a constrained controller—can ask what should move, what should remain, and what every edit changes in return.
System boundary
Image evidence first.
In the implemented workflow, ISR builds evidence fields from luminance, gradient, tone, and chromatic activity. VTL reads spatial and temporal coordinates from those fields. The Color Kernel adds color, palette, value, and color–structure relationships. The rendered image—not the camera metadata—is the evidence.
Input controlscamera · blocking · light · material
→RendererBlender or Unreal image
→Evidencefields · masks · regions
→VTLcoordinate vector + change
→Decisionretain · compare · revise
Core spatial kernel
Eight useful families. None is a ruler of quality.
rᵥ
Void
Low-activity or non-mask proportion in a named evidence field.
Not one universal measure of semantic negative space.
ρᵣ
Packing
Active structural pixels relative to their convex-hull area.
Not physical crowding or direct adjacency between story objects.
μ
Cohesion
Largest connected component’s share of the active structural mask.
Not complete compositional coherence or proof of intention.
xₚ
Peripheral pull
Share of masked structural weight inside the outer frame band.
A scalar boundary engagement measure, not a directional force vector.
θ
Orientation stability
Concentration among structural orientation families.
Not necessarily gravity alignment or a preferred absolute angle.
dₛ
Structural thickness
Skeleton-to-boundary width of the 2D structural mask.
Not physical depth, material thickness, or layering in 3D.
SDI
Dispersion
Mass-weighted spread around the field centroid.
Not organization, value, or disorder on its own.
The exact formulas, thresholds, bands, and normalizations are implementation-specific. A shared symbol across historical notebooks and the adapted observatory kernel does not guarantee numerically interchangeable values.
G channels and composites
Different questions require different lenses.
Scale-space center path
Tracks how a structural or intensity centroid changes as one image is represented at several blur scales. Structural G1 includes reliability gates. It is not eye tracking and blur scale is not viewing time.
Void-region geometry
Reads thresholded void coverage, component count, elongation, and the largest region. It is a proxy tied to the chosen field—not direct semantic figure/ground understanding.
Contour behavior
Measures turning-angle variance and an inflection proxy along detected contours. It differs from G1 curvature and from colloquial artistic “torque.”
Orientation entropy
Measures the breadth of edge orientations. It does not detect T-junctions and should not be described as literal depth or occlusion understanding.
Visual-load composite
The latest full-readout lineage combines capped G1–G4 contributions into a 0–5 composite. It is useful as one view of load, but caps can saturate and equal stated weights need not create equal effective influence.
Structural-coherence composite
Combines regional consistency, orientation alignment, cross-scale stationarity, and rhythm. High SCI is not proof that a decision was deliberate, resolved, or artistically successful.
Color, value, and directional fields
Color is measured as structure, not decoration.
The Color Kernel keeps lightness, chroma, hue, and edge structure available as related but separable evidence. This lets a brief move a color relationship while protecting the frame’s broader geometry—or expose when that separation is impossible.
Lightness and chroma distributions
Means, spread, and percentiles describe how bright and chromatically intense the field is. Light and dark mass fractions, contrast, and their centroids describe where value gathers.
Hue and palette organization
Hue entropy, chroma entropy, dominant-hue occupancy, palette compression, effective palette count, and palette weights describe concentration and diversity—not harmony or taste.
Chromatic placement
Chroma centroid offsets, focality, and color dispersion report where chromatic activity sits and how tightly it concentrates relative to the frame.
Directional energy
Horizontal, vertical, diagonal, and orthogonal shares describe the orientation carried by visible structure. They can help test directional pressure without naming its dramatic meaning.
Figure, ground, border, and focality
Figure–ground contrast, negative-space fraction, border salience, balance, focal count, focal dominance, spread, and centrality pressure describe how attention-bearing evidence is distributed.
Color–structure coupling
Mass–color alignment, edge–color coupling, chroma-on-structure, and light/dark alignment test whether chromatic and structural organizations reinforce or separate from one another.
Important distinction
Color independence summarizes alignment gaps and the difference between mean luminance and mean chroma fields. It is not the same quantity as ISR chromatic-boundary independence, and it is not a claim that color is perceptually or artistically independent of form.
Extended evidence families
The full readout is wider than one score.
Not every family is active or dependable in every run. The instrument records availability, thresholds, and provenance so a missing, saturated, or unstable channel cannot quietly become evidence.
Tonal construction
Bright/dark regions, negative-space structure, construction survival across thresholds, and self-similarity test how the frame’s value architecture persists.
Modeled salience paths
Focal concentration and simulated scan paths offer a reproducible attention model. They are computational probes, not recorded human gaze or viewing time.
Islands and relationships
Connected regions, accents, adjacency, overlap, gaps, and dependency describe the internal organization of masks without assigning semantic identities.
Density and resistance
Layer 6 families test local density, resistance, drains, and related field behavior when the needed masks and reliability conditions are present.
Sequence and intervention
Ranges, total variation, reversals, crossings, drift, and before/after deltas describe how coordinates move through time or respond to a controlled edit.
Evidence quality and availability
Render validity, mask coverage, sample count, reliability gates, repeat floors, and implementation versions determine whether a number is usable at all.
Prompt Pressure, Compositional Predictability, Sequence Drift Lock, Inversion Drift Check, and Symbolic Gravity Flags remain proposed validators in the supplied conceptual scaffold; they are not presented here as established always-on measurements.
Choice or defaults
The finished image cannot prove intention.
A centered image may be chosen. An accident may be recognized and retained. A complex image may be automatic. VTL can show structural relationships; process evidence is required to distinguish a documented choice from an inherited default.
What is measured?
Describe the image’s relationships without pretending they reveal motive.
What is known about process?
Record starting settings, alternatives, interventions, and accepted decisions.
What does this intention request?
Decide which relationships to target, protect, observe, or leave free.
VTL does not need to prove that a decision was made. It should make the decision available to be seen, retained, questioned, or changed.
The center is the delta
There is more than one center—and more than one delta.
| Delta | What changes | What it does not establish |
|---|---|---|
| Placement offset | A field centroid relative to the frame center | Whether centering was chosen |
| Scale-space displacement | A centroid across representations of one frame | An actual eye movement or viewing duration |
| Temporal displacement | A coordinate across successive frames | The absolute tension of either frame |
| Intervention delta | A coordinate before and after a controlled edit | Independent control of all other coordinates |
| Profile-score delta | A result under different reference weights | A change in the underlying image |
Operational grammar
Every channel receives a role for a particular brief.
Move toward a declared value, range, direction, or trajectory.
Remain within an explicit tolerance while something else changes.
Report the consequence without using it to select candidates.
Allow movement, but disclose relevant side effects rather than hiding them.
This is the practical distinction between an instrument and a universal score. The same high value may be requested in one shot, restrained in another, or simply watched in a third.
See these roles operating in the experiments