Overview

The Spatial Composition Framework (T19) is the territory of operations that take a spatial composition — a painting, a garden, a room, a page, a film frame, a dashboard, an urban scene, a network diagram considered as image — and analyze what the spatial structure itself does as primary content. The voids, the groupings, the force vectors, the affordances, the information density. The framework is not about what the elements depict; it is about what the layout does. The distinction matters because every spatial artifact has two readings available: what the parts say (which T11 handles) and what the arrangement of the parts itself communicates (which T19 handles). When you ask whether a dashboard is well designed, you are asking a T19 question. When you ask what the dashboard is reporting, you are asking a T11 question. The same artifact, two different operations.

The framework runs in four modes. Ma Reading is the contemplative reading of the void, the interval, the silence — drawn from Japanese aesthetics (Ma, Yūgen, Wabi-sabi, Mu) and extended to temporal compositions through Cage and Bordwell. The mode identifies operative voids (only what is load-bearing, not all empty space), names what each void does in tradition-specific vocabulary, performs the removal/alteration test (would replacing the void with content of equal weight alter the work?), and traces suggestion-resonances. Compositional Dynamics is the universal-perceptual reading drawn from Gestalt psychology and Arnheim — predicts how perception parses the visual field via grouping principles, identifies the structural skeleton (axes, center, frame), assigns visual weights on empirical grounds (size, contrast, color, isolation, position, depth), names force vectors and tensions, classifies dynamic equilibrium, predicts the eye-path. Place Reading and Genius Loci is the deep place-reading drawn from Alexander, Norberg-Schulz, Lynch, Bachelard, Appleton, and the Kaplans — integrates six analytical operations on inhabited or inhabitable space (prospect-refuge balance, active pattern-language patterns, Lynchian legibility, restorative properties, genius loci character-of-place, Bachelardian topoanalysis) and produces predictions of inhabitation grounded in spatial features. Information Density is the applied analytical reading on information graphics drawn from Tufte, Bertin, Cleveland-McGill, and Bringhurst — audits data-ink, checks visual-variable mapping for fitness, checks elementary-perceptual-task fitness, analyzes typographic hierarchy, and produces ranked specific recommendations.

The framework’s load-bearing intellectual content is the distinction between layout-as-meaning and layout-as-notation, the operative-voids discipline, the empirical visual-weight requirement, and the affordance-grounding rule. The layout-as-meaning vs. layout-as-notation distinction is what gives T19 its territory boundary against T11. The operative-voids discipline is Ma Reading’s central methodology — not all empty space is load-bearing; only the voids that, if filled with equal-weight content, would alter the work substantively count as operative. The empirical visual-weight requirement is Compositional Dynamics’ guard against symbolic-weight confusion — assignments must rest on size, contrast, color, isolation, position, and depth, not on what the depicted thing means. The affordance-grounding rule is Place Reading’s guard against analyst-projection — proposed affordances must be grounded in features of the space, must survive an inhabitant of different stature/ability/culture, and must produce testable behavioral predictions rather than sentiment statements.

The framework deliberately resists four patterns that fail compositional reading. Surface description as compositional analysis — counteracted by the perceptual-parse-with-cues requirement in Compositional Dynamics. Aesthetic projection without spatial warrant — counteracted by Place Reading’s affordance-grounding rule. Chartjunk-as-label without specific identification — counteracted by Information Density’s mark-by-mark data-ink audit. Concept-melt across Japanese aesthetics traditions — counteracted by Ma Reading’s named-tradition discipline (Ma, Yūgen, Wabi-sabi, and Mu are not interchangeable).

The framework answers questions like: Read this painting compositionally — what is the layout doing? What is the silence doing in this scene? Will this room feel restorative or depleting? Is this dashboard’s visual encoding doing the right job? What does this chart’s encoding ask the reader’s perception to do, and is that the right ask for the message?

Systemic context

Spatial Composition is a Tier-3 cognitive territory in the Ora reasoning lattice, paired with Structural Relationship Mapping (T11) at the canonical T11↔T19 disambiguator — the same diagrammatic input answers different questions depending on whether the user wants the relations the diagram asserts (T11) or what the layout itself is doing (T19). When both legitimately fire, T11 runs first because relation-extraction is the lighter, more determinate operation; T19 layers compositional reading on top. T19 also adjacent to T20 (Open Exploration) when the input is aesthetic and the prompt is broad — analytical reading routes to T19, open exploration to T20, and both may fire on the same input. The four modes within T19 escalate sideways to one another: Compositional Dynamics is the universal-perceptual default when the input does not signal aesthetic-vs-applied clearly; Ma Reading hooks for the contemplative complement on aesthetic input; Place Reading hooks when the question becomes what this place is; Information Density hooks when chart-encoding-misfit is the specific question rather than generic compositional critique. A fifth mode (Information-Graphic Visual-Hierarchy Analysis) is held in reserve with an explicit promotion threshold of ~15% T19 invocation share or visible failure of Compositional Dynamics to distinguish encoding-misfit from generic compositional critique.

Ora’s output

Example question: Read this dashboard for me compositionally. It’s a financial-operations dashboard with twelve KPI tiles in a 4×3 grid, three line charts below, and a sidebar with filters. The team uses it daily but says it “feels off.” What is the layout doing?

Ora’s response (running T19 in Compositional Dynamics, with sideways hook to Information Density):

The input is an applied composition and the question is what the layout is doing — Compositional Dynamics is the right default. The “feels off” complaint is the kind of report compositional analysis can diagnose: the perceptual parse is producing an experience the team can detect but not name.

Perceptual parse and figure-ground. The 4×3 KPI grid reads as a single figure (Gestalt: common region + similarity of tile shape + grid regularity). Within the grid, tiles group secondarily by color saturation — the four highest-revenue tiles use a saturated green, the others a desaturated grey-green. The line charts read as a separate figure; the sidebar reads as ground (low saturation, peripheral position). The perceptual hierarchy puts KPIs on top, charts below, filters in the periphery.

Structural skeleton. The dashboard’s bounding rectangle puts its geometric center at the seam between row 2 and row 3 of the KPI grid — which is occupied by the seventh and eighth tiles. Those two tiles are not the most important (the team confirms revenue and cash-on-hand are load-bearing and are tiles 1 and 2, top-left). The structural skeleton is putting weight on positions the data doesn’t warrant. This is one source of the “feels off” — the eye is pulled to the center, but the center is not where the meaning lives.

Visual weight per element. Tiles 1 and 2 carry the saturated green and the largest numeric type — high visual weight. Tiles 7 and 8 carry medium saturation and medium type but the structural-skeleton position amplifies their weight beyond their content’s importance. Tile 12 (bottom-right, “team headcount”) uses an icon-plus-number encoding that breaks the grid’s similarity — it reads as anomalous, which the eye treats as either important or as error.

Force vectors and dynamic equilibrium. The grid’s regularity creates stable equilibrium. The saturated-green cluster (tiles 1, 2, 5, 6) creates a leftward pull. The anomalous tile 12 creates a rightward-downward pull. The line charts pull the eye downward. The net field is unbalanced — the eye starts at top-left, drifts down to charts, bounces to bottom-right, back up. The eye-path does not honor the team’s actual reading priority. This is the second source of “feels off.”

Hook to Information Density. Tile 12 is a Bertin misfit — quantitative data on an icon-plus-number combination where the icon carries no data (chartjunk per Tufte); recommendation is the same numeric-tile encoding as the other eleven. The line charts use color-by-series with six series — Cleveland-McGill rank says color-hue is among the lowest-accuracy encodings for distinguishing series; recommendation is small-multiples using position-on-common-scale (rank-1 task).

Recommendations ranked by impact. (1) Move tiles 1 and 2 to the geometric center — align structural skeleton with content priority; high impact. (2) Replace tile 12’s encoding with the standard numeric tile; medium-high impact. (3) Convert line charts to small-multiples; medium impact. (4) Decide whether the saturated-green cluster is intentional grouping or accidental — if intentional, strengthen with a shared border; if accidental, desaturate tiles 5 and 6.

Confidence per finding. High on (1) and (2). Medium on (3) — small-multiples helps for cross-series comparison; if usage is single-series tracking, the current encoding is fine. Medium on (4) — depends on team intent.

The “feels off” report decomposed into three identifiable mismatches between perceptual structure and content priority; each carries a specific recommendation grounded in compositional and information-density principles; the team can act on each separately.

Commercial AI comparison

Comparison content auto-populates when the comparison-refresh framework runs against this question. Drafters do not author this section.

Brief comparison commentary

Auto-populates with the comparison content above.

How to use this framework

You can run the Spatial Composition pattern with any AI of your choice. The composition is single-pass against any of the four modes.

The prompt:

[Paste the framework specification.]

[Optional: state mode explicitly — Ma Reading, Compositional Dynamics, Place Reading, or Information Density — or let the framework auto-route via the within-territory disambiguation question.]

Composition: [Paste image, link, or detailed verbal description. For Compositional Dynamics on visual input, image fidelity matters — high-resolution photos or screenshots work; rough sketches degrade the perceptual-parse step.]

Question (optional): [What you want to learn from the composition.]

The AI runs the within-territory disambiguation, picks the mode, and produces the mode-specific output. For Compositional Dynamics, expect eight required sections (perceptual parse, structural skeleton, visual weights, force vectors, dynamic equilibrium, eye-path, ambiguity loci, confidence). For Ma Reading, expect five (operative voids, what each does, what would collapse without it, suggestion resonances, counter-readings). For Place Reading, expect ten (place summary and scale, prospect-refuge balance, active pattern-language patterns, Lynchian legibility, restorative properties, genius loci, Bachelardian topoanalysis, predicted inhabitation, design recommendations, counter-readings). For Information Density, expect nine (graphic summary, data-ink audit, variable-fitness check, perceptual-task fitness, typography, chartjunk inventory, ranked recommendations, residual tradeoffs, confidence per recommendation).

For best results:

  1. Provide image input where possible. Verbal description works for high-fidelity descriptions but degrades for rough sketches. Compositional Dynamics in particular benefits from actual visual input rather than verbal proxy.
  2. State the question. “Read this composition” is acceptable; “what is the silence in this scene doing?” or “is this dashboard’s encoding doing the right job?” routes more precisely.
  3. Don’t expect Information Density on aesthetic inputs. A painting is not a dashboard; chart-encoding analysis is the wrong tool. The disambiguation routes accordingly, but if the AI is forced to apply Information Density to a Rothko, the output will be uninteresting.
  4. Honor the mode boundary. When the question is what the relations in a diagram assert rather than what the layout is doing, the AI should route to T11 (Structural Relationship Mapping) rather than producing a T19 reading of a notational artifact.

The framework is deliberately tool-agnostic. The four modes’ analytical disciplines, the operative-voids test, the empirical visual-weight requirement, and the affordance-grounding rule are conceptual disciplines that survive the lift to any environment. The output is plain markdown with embedded image annotations where applicable.

Other examples

  • Ma Reading on a Hiroshi Sugimoto seascape — Identifies the void as load-bearing (absence of incident is the content); names what it does (Ma generates duration; Yūgen invites contemplation; Wabi-sabi reads as patina-of-time); performs the removal test (any minimal incident would destroy the work); offers counter-reading for viewers expecting figurative content. Demonstrates contemplative-articulative mode where the void is the work.

  • Place Reading on a planned office redesign — Six-cluster pass on floor plans: prospect without refuge (concerning for sustained focus); pattern-language violations (Light on Two Sides, Alcoves absent); Lynchian legibility weak (no landmarks for wayfinding); low restorative properties; genius loci reads as transient. Predicts inhabitation: focused work declines; staff begins informal redesign within six months. Demonstrates affordance grounding producing actionable critique.

  • Compositional Dynamics on a film still — A Wes Anderson still: extreme bilateral symmetry, flat picture plane, evenly distributed weight, centered figure, saturated palette with limited hue range. Force-vector reading: symmetry creates stable equilibrium with no directional pull; lack of depth flattens scene into tableau. The “Anderson feel” decomposes into specific compositional choices. Demonstrates diagnosing stylistic signatures rather than critique.

Citations

The Spatial Composition Framework integrates four tradition clusters under one territory. Japanese aesthetics provides Ma Reading: Isozaki’s MA: Space-Time in Japan, Nitschke’s placement-vs-place distinction, Zeami’s Fūshikaden on Yūgen, Tanizaki’s In Praise of Shadows, Suzuki on Mu; Cage’s 4’33” and Bordwell’s poetics of cinema extend Ma to temporal compositions. Compositional Dynamics rests on Gestalt psychology (Wertheimer, Köhler, Koffka; Wagemans et al.’s 2012 century review) and Arnheim’s Art and Visual Perception and The Power of the Center. Place Reading integrates six clusters: Alexander’s A Pattern Language, Norberg-Schulz’s Genius Loci, Lynch’s The Image of the City, Bachelard’s The Poetics of Space, Appleton’s prospect-refuge habitat theory, and Kaplan’s attention restoration theory. Information Density draws on Tufte’s data-ink, Bertin’s Semiology of Graphics, Cleveland and McGill’s perceptual-task ranking, and Bringhurst’s Elements of Typographic Style.

The territory carries five open debates at the territory level: the spatial-vs-compositional naming question; aesthetic-only-or-also-abstract scope; the Western-analytical-and-Eastern-aesthetic relationship; verbal accessibility for AI implementation; and mode granularity (whether more tradition-specific modes should be promoted). The framework is single-author and originated 2026-05-01 from the T19 reanalysis under Decision G; the renaming from “Visual and Spatial Structure” reflects the territory’s actual subject (composition-as-content, not relation-extraction — the latter is T11).

Downloads

  • Framework specification (PDF) — link to ora-ai.org canonical artifact when published
  • Framework specification (plain text) — link to ora-ai.org canonical artifact when published
  • Full white paper (PDF) — link when published