Deterministic Interaction and Accessibility Specifications for Spatio-Temporal Globe and VR Simulations

Full report

On this page

Introduction to Advanced Spatio-Temporal Visualization

The advent of highly complex intelligence and environmental monitoring programs, such as the Intelligence Advanced Research Projects Activity (IARPA) Space-based Machine Automated Recognition Technique (SMART) and the Time, Images, and Geo-location Retrieval (TIGeR) framework, necessitates revolutionary approaches to spatio-temporal data visualization1. These programs aggregate vast repositories of multimodal data—ranging from satellite imagery and geolocation metrics to interconnected intelligence reports—requiring analysts to navigate dynamic, interconnected nodes across massive spatial and temporal scales3. The cognitive overhead involved in tracking these entities is immense, particularly when attempting to harmonize spatial, spectral, and temporal data to identify anthropogenic activities, structural developments, or socio-political movements2. When integrating these architectures into Virtual Reality (VR) or advanced globe-view web simulations powered by WebGL or similar rendering APIs, a fundamental engineering and user-experience challenge arises. The system must maintain the cognitive and visual continuity of selected events, places, people, institutions, and reports as the user manipulates the view5. Random, heuristic-based, or non-deterministic visual transformations lead to chaotic spatial layouts, causing immediate context loss, undermining spatial orientation, and ultimately hindering intelligence analysis8. To solve this, rendering engines must implement a rigorous, deterministic interaction specification. This specification must govern how complex spatial graphs behave when users pan, zoom, rotate, resize, change orientation, or transition between three-dimensional globes and two-dimensional or volumetric temporal timelines10. Furthermore, the deployment of such immersive platforms within government or enterprise ecosystems mandates strict adherence to accessibility standards. These platforms must remain fundamentally accessible, complying with W3C Web Accessibility Initiative \- Accessible Rich Internet Applications (WAI-ARIA) guidelines via sophisticated Shadow Document Object Model (DOM) implementations and spatial keyboard routing algorithms12. The ensuing analysis details an exhaustive, deterministic interaction specification engineered to orchestrate entity connectivity, label placement, occlusion handling, layout safety, context-loss recovery, and comprehensive accessibility features for state-of-the-art interactive globe and timeline simulations.

Entity Connectivity and Spatio-Temporal Data Topologies

Before establishing the algorithmic rules for visual rendering, the underlying data topology of the intelligence platform must be rigorously defined. In advanced spatial computing and visual analytics platforms, entities are not merely isolated geometric points; they act as interdependent nodes within a highly interconnected, multi-relational knowledge graph5. The spatial graph maps complex real-world concepts into a digital, manipulable framework. Within the context of the IARPA SMART and similar programs, the data model inherently comprises five primary entity types that must be visually tracked and managed. Events represent temporal occurrences bounded to specific coordinate ranges and timeframes, such as localized heavy construction phases or political protests5. Places define static or fluid geographic bounding boxes and coordinates, ranging from distinct cities to disputed maritime zones and shifting environmental borders. People denote individual actors or demographic cohorts whose locations dynamically shift across the time axis, forming complex spatial trajectories. Institutions represent organizational entities that frequently exhibit decentralized or multiple spatial footprints, complicating their anchor points on a globe. Finally, Reports comprise unstructured or semi-structured multimedia and text files that serve as the evidentiary foundation, semantically linked to the aforementioned entities10. To maintain connectivity during high-velocity navigation, the simulation must rely on a directed, multi-relational graph structure. The visual representation of this graph utilizes geometric nodes to represent the entities and connector lines, or edges, to represent their semantic or temporal relationships. The persistence of these connections must remain visually coherent and mathematically predictable regardless of the user's viewport manipulation10. When an analyst observes a linkage between a person of interest and an institution, that visual linkage must not break, tangle, or disappear arbitrarily during a zoom or rotation operation. This requires a robust geometric foundation that recalculates edge routing in real-time, treating the graph not as a static overlay, but as a physical system of springs and constraints embedded within the virtual environment19.

Topological Morphing: Transitioning Between Globe and Timeline Views

A critical feature of the specified simulation is the ability to seamlessly move between a spatial 3D globe view and a temporal timeline view. In traditional geographic information systems, switching from a map to a timeline involves a hard visual cut, destroying the user's configural knowledge and forcing a complete cognitive reset21. To preserve entity relationships and support continuous storytelling, the transition must rely on a topological morphing strategy, specifically the implementation of a Space-Time Cube (STC)11. The Space-Time Cube conceptualizes data within a four-dimensional hypercube, which is projected down to three dimensions for rendering. When the user triggers a transition from the globe to the timeline, the spherical projection of the globe deterministically flattens into a planar geographic map, forming the [Figure omitted from source export] and [Figure omitted from source export] axes of the new view11. Concurrently, the timeline extrudes along the [Figure omitted from source export]\-axis, representing depth or height, creating a three-dimensional volume where events are stacked chronologically22. Entities that share a geographic location but occur at discrete temporal intervals are stretched along this [Figure omitted from source export]\-axis, connected by vertical or near-vertical trajectory splines that visualize their chronological progression. The camera interpolation during this transition must be strictly deterministic. The engine calculates a continuous bezier curve for the camera's translation and a spherical linear interpolation (slerp) for its rotation, moving smoothly from a geocentric orbit to an orthographic or perspective view of the STC. To mitigate the visual complexity of dense data clusters during this volumetric transition, the engine must employ "micro time steps." Rather than displaying all temporal data simultaneously, which results in a monolithic block of overlapping nodes, micro time steps dynamically expand the temporal gaps or filter intermediate data points, ensuring that the transition maintains a fluid, readable density23. This unified geo-temporal representation has been proven to significantly increase an analyst's understanding of entity behaviors and relationships, facilitating the perception of the "who" and "what" within the context of the "where" and "when"10.

Deterministic Callout Placement and Force-Directed Algorithms

The crux of the rendering challenge in 3D spatial computing lies in managing how textual labels and geometric callouts behave when the user manipulates the environment. Label placement in 3D environments chronically suffers from overlapping, causing severe visual clutter and rendering textual data illegible9. While static 2D cartography can utilize greedy algorithms or pre-computed bounding boxes, a 3D VR or globe environment requires dynamic, continuous updating9. This specification mandates a mixed-labeling approach that operates in screen space, utilizing a force-directed layout algorithm combined with fuzzy multi-criteria decision making, such as the Beams Displacement Method25. In this framework, callouts function as external labels connected to their geographical anchors via leader lines. To determine their precise screen-space coordinates dynamically during a camera transformation—such as a pan or rotate operation—the simulation computes a physics-based equilibrium of simulated forces at every frame20. This treats the user interface as a mechanical system where callouts act as repelling magnetic nodes and leader lines act as tethering springs19.

The physics engine calculates the total force vector [Figure omitted from source export] for each label in the scene. The callouts are iteratively shifted by a damped velocity vector until the system reaches a state of local energy minima19. Because the initial seed positions are derived deterministically from the anchor coordinates, and the algorithm relies on fixed tuning parameters ([Figure omitted from source export]), the resulting layout is inherently stable and reproducible. If an analyst shares a specific viewport configuration with a colleague, the force-directed algorithm will resolve to the exact same visual layout on the colleague's machine, fulfilling the requirement for a deterministic interaction specification8.

Connector-Line Origin, Termination, and Geometric Routing

Connector lines, frequently referred to as leader lines in cartographic contexts, establish the vital visual relationship between a textual callout floating in screen space and a geographical point anchored in world space on the 3D globe9. The management of these lines is paramount; poorly routed lines introduce visual noise that destroys the utility of the spatial graph. To ensure lines do not obscure the text they annotate or cut discordantly through user interface elements, the origin and termination points must adapt dynamically based on the callout's relative position to the geographical anchor. The termination point is fixed; the leader line must terminate precisely at the 3D world coordinate of the event or place, translated into the active camera frustum. However, the origin point on the callout end must be dynamic. The bounding box of the callout is assigned eight potential origin anchor points: top-left, top-center, top-right, right-center, bottom-right, bottom-center, bottom-left, and left-center. The geometric routing algorithm continuously calculates the Euclidean distance from the screen-space projection of the world anchor to all eight of these bounding box points. The point yielding the shortest distance—without its path intersecting the interior of the bounding box itself—is deterministically selected as the active origin. As a label floats around its anchor due to the force-directed physics simulation, the leader line origin smoothly snaps to the most optimal edge, preventing the line from striking through the text. In instances where entities themselves are connected directly to one another—such as tracing a person's movement from one geographic coordinate to another, or mapping financial transactions between decentralized institutions—standard straight lines are geometrically problematic on a spherical globe, as they would clip through the earth's crust. To resolve this, the rendering engine utilizes 3D Bezier curves. The curvature radius of these splines is computed proportionally to the geodesic distance between the source and target nodes on the globe's surface. The curves arch outward along the normal vector of the midpoint between the two anchors, creating a semi-circular bridge that perfectly avoids intersecting the planetary geometry28. The height of the arch allows analysts to perceive long-distance connections instantly without rotating the globe, acting as an implicit visual indicator of the distance between the connected entities.

Occlusion Management, Backface Culling, and Transparency Auras

As the user rotates the globe or navigates through a dense 3D urban environment, entities will inevitably pass behind the curvature of the Earth or behind towering 3D structures. Attempting to render all spatial graph nodes continuously results in severe visual noise, rendering the application unusable. Consequently, rigorous occlusion management is required, utilizing a combination of geometric backface culling and sophisticated transparency auras9. The visibility of an anchor point on a spherical globe can be determined with high mathematical efficiency using the dot product of the camera's view vector and the surface normal at the anchor's specific location30. Let [Figure omitted from source export] be the normalized vector drawn from the anchor point to the camera lens, and [Figure omitted from source export] be the normalized surface normal vector of the globe radiating outward at the anchor point. If [Figure omitted from source export], the point is facing the camera and is theoretically visible. If [Figure omitted from source export], the point has rotated past the horizon and is strictly occluded. When an entity is determined to be occluded by the planetary horizon, its associated callout and connector line must be managed elegantly. Instantly hiding the label causes a visual popping effect that is highly jarring and disrupts the user's spatial tracking. Instead, the opacity of the callout is mapped linearly to the dot product result. As the anchor approaches the horizon ([Figure omitted from source export]), the callout and its leader line fade out smoothly, signaling to the user that the entity is slipping out of view. Foreground occlusion presents a different challenge. If a foreground 3D object, such as a volumetric building in a high-zoom city view, occludes an internal label attached to a smaller feature behind it, rendering the label strictly behind the building destroys analytical context9. In this scenario, the rendering pipeline employs a depth-aware shader that evaluates the Z-buffer. If the label's spatial coordinates are deeper than the geometry blocking it, a "transparency aura" or stencil halo is procedurally cut out of the foreground geometry. This allows the label to subtly shine through the occluding object. This technique preserves the spatial depth cues of the 3D world—the user implicitly understands the label is behind the building—while completely preserving the readability of the critical intelligence data9.

Viewport Dynamics: Safe Areas, Boundary Constraints, and Resizing

Intelligence platforms are rarely viewed in pristine, full-screen isolation; users frequently resize browser windows, open side panels, and snap applications to different monitor quadrants. When a user resizes the simulation viewport, the spatial graph must instantly adapt to prevent critical entities from being pushed off-screen into invisible rendering buffers. To achieve this, the deterministic specification enforces structural "safe areas." These safe areas establish a rigid padding perimeter—for example, a boundary inset 50 pixels from all viewport edges—within which all floating callouts must be constrained31. The edges of this safe area are injected into the force-directed physics simulation as infinite mass planes. If a window resize event shrinks the viewport, the physics algorithm detects the new boundary constraints in real-time. Repulsive forces emanating from the invisible "walls" of the safe area actively push external labels inward toward the center of the screen. If the viewport is shrunk significantly, the density of callouts forced into the remaining safe area will inevitably exceed a predefined readability threshold, leading to a catastrophic collapse of the layout where labels overlap despite repulsive forces. To prevent this, the engine monitors the bounding-box density ratio. Once the ratio of label area to available screen area crosses the threshold, the system deterministically triggers a Semantic Zoom transition, actively collapsing granular data points into summarized, high-level clusters to restore legibility.

High Zoom and Semantic Level of Detail (LoD)

Navigating a massive spatial database requires moving seamlessly from a macro-planetary perspective down to a micro-street level. As a user zooms out (increasing altitude) or zooms in (decreasing altitude), relying solely on geometric scaling—where 3D objects and labels simply become larger or smaller—is a fundamental failure of visualization32. If geometric zoom is strictly applied, zooming out would cause thousands of overlapping callouts to crowd the screen, while zooming in would result in sprawling, empty expanses of geometry. To resolve this, the specification mandates Semantic Zooming paired with hierarchical clustering algorithms. Semantic zooming alters the fundamental type, meaning, and granularity of information displayed based on the camera's altitude, rather than merely adjusting its geometric scale32. The deterministic interaction is governed by predefined altitude thresholds integrated into a Level of Detail (LoD) tree data structure32.

The transition between these LoD states must be deterministic and visually continuous. When an aggregate node breaks apart during a zoom-in operation, the child nodes animate radially outward from the parent node's origin, settling into their true geographic positions. This animation preserves the user's configural understanding of how the macro-data relates to the micro-data, bridging the gap between broad statistical overviews and granular investigative analysis.

Navigation Stability and Context-Loss Recovery Strategies

Immersive spatio-temporal environments present a high cognitive risk of user disorientation. When navigating highly complex spatial graphs across a full 3D globe, users can easily "lose" their subject matter if they rotate the earth too quickly, zoom in on an empty region of the ocean, or lose track of an entity transitioning across the timeline.

Handling High-Velocity Pan, Zoom, and Rotation

During rapid user inputs—such as an aggressive mouse drag to spin the globe to the opposite hemisphere—running the full force-directed label placement algorithm at 60 or 120 frames per second can be computationally prohibitive. More importantly, it can cause labels to jitter violently as they constantly fight shifting spatial constraints. To smooth high-velocity interactions, the system implements an interaction state machine. During active panning or rotation, the callouts temporarily detach from the physics simulation. They switch to strict, static screen-space offsets relative to their world-space anchors, moving exactly in tandem with the underlying geography, while leader lines remain locked in their last known optimal configuration. Upon input cessation (when the user releases the mouse button or stops scrolling), the physics algorithm is immediately re-engaged. The callouts gracefully spring into their newly optimized positions based on the updated camera angle, resolving any new occlusions or overlaps that occurred during the rotation.

Deterministic Context-Loss Recovery

If a user pans away from an active selection of events, places, or reports, the system must provide robust mechanisms to recover the context without forcing the user to manually hunt for the data across the globe. The primary mechanism for this is the implementation of off-screen indicators. When an actively selected entity moves outside the camera's viewing frustum (the geometric bounds of the viewport), boundary labels are automatically generated. These indicators are anchored to the extreme edges of the screen's safe area34. The specification supports three distinct modes of off-screen labeling to provide maximal context:

1. Angle-Encoded Indicators: A geometric chevron or arrow points directly toward the shortest geodesic path to the off-screen object, providing instant directional awareness34.

2. Height-Encoded Indicators: Crucial for timeline or Space-Time Cube views, these indicators rest at the exact [Figure omitted from source export]\-axis elevation on the edge of the screen corresponding to the off-screen object. This allows the user to instantly understand the object's chronological position relative to their current view34.

3. Value-Encoded Indicators: The indicator displays a highly summarized data point alongside the arrow (e.g., "Target Alpha: 2,000 km away"), allowing the analyst to weigh the importance of navigating back to it34.

Furthermore, the system maintains a contextual spatial breadcrumb trail, tracking the camera's historical waypoints. A dedicated "Recenter Context" function executes a deterministic interpolation of the camera transform matrix, flying the viewport back to a mathematically calculated bounding sphere that perfectly encapsulates all currently active and selected entities in the spatial graph.

Comprehensive Accessibility Architecture: ARIA and the Shadow DOM

Providing deep, immersive interaction is fundamentally insufficient if the visualization excludes analysts who rely on assistive technologies (AT), such as screen readers, braille displays, or advanced keyboard navigation. WebGL and canvas-based 3D rendering engines are inherently opaque to the browser's Document Object Model (DOM). To a screen reader, a deeply complex IARPA globe simulation appears simply as a blank, meaningless \<canvas\> tag14. The deterministic specification must construct an intricate accessibility layer to bridge the vast gap between the visual 3D representation and the browser's internal Accessibility Tree.

The Shadow DOM and Accessibility Tree Flattening

To render the 3D globe accessible, the system utilizes a parallel "Shadow DOM" architecture. For every semantically meaningful 3D node rendered in WebGL (every event, place, person, and report), a corresponding semantic HTML element is dynamically generated. These elements are placed inside a visually hidden container, or natively within a closed Shadow DOM root attached to the primary web component13. When the web browser processes this structure, it executes a vital operation known as flattening. The browser merges the main, visible Light DOM and the hidden contents of the Shadow DOM into a single, unified Accessibility Tree13.

• Screen readers (such as JAWS, NVDA, or Apple VoiceOver) do not interact with the pixels painted on the 3D canvas; they parse this flattened Accessibility Tree13.
• As the data state changes in the 3D globe—for example, zooming in triggers a Semantic Zoom that explodes a cluster into ten individual nodes—the JavaScript engine synchronously constructs ten new HTML elements in the hidden Shadow DOM38.
• Nodes that are visually culled by the rendering engine (e.g., hidden behind the earth via backface culling) are either dynamically removed from the Shadow DOM entirely, or explicitly marked with the aria-hidden="true" attribute. This strict synchronization ensures the screen reader user only processes what is currently "visible" in the viewport, preventing cognitive overload from thousands of hidden data points35.

WAI-ARIA Graphics Module Integration

Standard HTML roles, such as generic buttons or links, are woefully insufficient for conveying the complex, multi-dimensional relationships inherent in spatio-temporal intelligence analysis39. To resolve this semantic deficit, this specification mandates the implementation of the W3C WAI-ARIA Graphics Module, which introduces domain-specific roles explicitly designed for complex graphical structures and charts12.

By aggressively utilizing the aria-describedby attribute, the application links a basic node in the Shadow DOM to extensive, off-screen metadata tables. This allows a screen reader user focusing on a simple point on the globe to seamlessly command their AT to read the full intelligence report associated with that coordinate, achieving total data parity with sighted users40.

Spatial Keyboard Navigation and Graph Routing

In a standard text-based webpage, pressing the Tab key moves focus sequentially down the DOM tree. In a 3D globe simulation, linear tabbing is functionally meaningless; an event occurring in New York has no logical sequential relationship to an event in London unless explicitly defined by an active query. Forcing a keyboard user to tab linearly through 5,000 global data points destroys usability. To achieve robust, intuitive keyboard navigation, the specification implements intelligent spatial graph routing.

1. Geodesic Shortest Path: The system continuously maps the visible nodes on the globe into a 2D proximity matrix based on their current screen-space coordinates. It calculates the geodesic representation—the shortest path across the curved surface of the earth45.

2. Directional Arrow Navigation: The user can navigate the spatial graph physically using the keyboard's directional arrow keys (Up, Down, Left, Right). When a specific node is in focus, pressing Right triggers a routing algorithm that calculates the nearest unoccluded neighbor node in the positive [Figure omitted from source export] direction of the screen space, transferring focus to it instantly.

3. Semantic Navigation: Alternatively, users can toggle a "Semantic Routing" mode. In this state, pressing Tab follows the timeline sequentially (e.g., jumping from an event in 2024 directly to the next chronologically related event in 2025, entirely regardless of their geographic distance on the globe).

4. Visual Focus Indication: When a hidden node receives keyboard focus in the Shadow DOM, the WebGL engine intercepts the focus event and visually highlights the corresponding 3D node with an accessible bounding box. It draws a stark, high-contrast vector line from the camera lens to the node, and the camera smoothly pans to center the focused node in the viewport, ensuring that sighted keyboard users (such as those with motor impairments) can track their location perfectly.

RTL (Right-to-Left) Layout Support in 3D Spaces

Standard internationalization for Left-to-Right (LTR) languages versus Right-to-Left (RTL) languages (such as Arabic, Hebrew, or Farsi) primarily involves aligning text blocks and mirroring web layouts. However, in spatial computing, timeline visualization, and VR, implementing RTL support fundamentally alters the user's cognitive spatial mapping and requires complex matrix transformations. For textual callouts, all HTML and CSS applied to the overlays—whether rendered via standard HTML elements superimposed over the canvas or Canvas 2D text generation—must dynamically respect the dir="rtl" attribute. Text alignment defaults to right-justified, and directional punctuation (such as periods at the end of Arabic sentences or brackets around coordinates) must render correctly through Unicode bidirectional algorithms without breaking the visual string logic. The deeper challenge arises during temporal transformations. When the system transitions from the geographic globe to the Space-Time Cube or flat timeline view, the temporal axis must fundamentally invert to support RTL cognitive models. In standard LTR environments, time conventionally flows from left to right, with the past on the left and the future on the right. In an RTL localized mode, this temporal layout matrix must be mathematically mirrored:

• The earliest chronological dates are positioned on the far right of the viewport, progressing toward the left.
• The camera's default sweeping animation during chronological playback reverses its vector, traveling strictly along the negative [Figure omitted from source export] axis.
• Off-screen indicators pointing to "future events" point toward the left edge of the screen instead of the right edge.

Critically, the geographical representation (the globe itself) is not mirrored. Geographic coordinates, longitude, and latitude are absolute physical constants. Attempting to mirror a map horizontally renders the geography unrecognizable and useless for intelligence purposes. Therefore, the rendering engine must utilize a bifurcated coordinate transformation matrix. It must preserve the absolute geospatial rendering of the earth, while selectively applying scalar inversions strictly to time-derived UI elements and the timeline's fundamental spatial axis. This hybrid layout approach guarantees cognitive harmony for RTL users analyzing absolute geographic data over time.

System Interactivity State Machine Integration

To successfully synthesize all the aforementioned rules—force-directed physics, backface culling, semantic zooming, accessibility synchronization, and spatial routing—the deterministic interaction specification must function as a strict, cyclical state machine. This architecture ensures that incredibly complex, high-volume IARPA-level queries do not destabilize the VR or Globe environment.

1. Idle State (Equilibrium): The physics simulation has reached local minima. Callouts are stable and highly legible. The hidden Shadow DOM is fully synchronized with the visual state.

2. Input Detected (Pan/Zoom): The physics engine instantly detaches labels, reverting them to rigid screen offsets. Shadow DOM updates are temporarily suspended to preserve CPU cycles and prevent the screen reader from queuing thousands of rapid changes. LoD threshold checks continuously monitor camera altitude.

3. Threshold Crossed (Semantic Zoom): Node clusters collapse into aggregates or explode into granular points. Transparency auras are recalculated against the Z-buffer.

4. Input Ceased (Resolution): The physics engine resumes. Repulsive and attractive forces resolve any new overlaps generated by the camera move. Occluded nodes are faded based on the dot product of the new camera vector. The Shadow DOM is completely re-rendered, and the browser's Accessibility Tree is updated.

5. Context Request: If the user initiates a search, triggers a contextual breadcrumb, or presses a directional navigation key, the geodesic navigation module overrides the camera matrix, deterministically interpolating the view to the new target and restarting the state cycle.

Conclusion

The visualization of multifaceted, spatio-temporal intelligence data—such as the complex correlations of events, places, people, institutions, and reports—demands far more than simply plotting arbitrary points on a rotating 3D sphere. To extract genuine, actionable value from programs like SMART and TIGeR, rendering engines must treat the interface as a dynamic, living graph bound by strict mathematical rules. By implementing a rigorous, deterministic interaction specification, advanced VR and globe simulations can transform chaotic raw data into cognitively stable narratives. Through the precise application of force-directed layout algorithms, backface culling dot-products, and mathematical occlusion management, visual clutter is definitively mitigated during even the most complex camera maneuvers9. Furthermore, by tightly synchronizing the 3D rendering pipeline with the WAI-ARIA Graphics Module and a dynamically updating Shadow DOM architecture, the simulation guarantees that these incredibly complex spatial graphs are entirely accessible to screen readers and keyboard-dependent analysts12. Ultimately, this specification ensures that as users pan, zoom, rotate, and transition through the dimensions of time and space, the integrity of the data topology remains perfectly intact, empowering uninterrupted discovery, collaborative intelligence synthesis, and deep analytical insight.

Works cited

1. TIGeR: A Unified Framework for Time, Images and Geo-location Retrieval \- arXiv, https://arxiv.org/html/2603.24749v2

2. TIGER: A Unified Framework for Time, Images and Geo-location Retrieval, https://openaccess.thecvf.com/content/CVPR2026/papers/Shatwell\_TIGER\_A\_Unified\_Framework\_for\_Time\_Images\_and\_Geo-location\_Retrieval\_CVPR\_2026\_paper.pdf

3. Automated Global-Scale Detection and Characterization of Anthropogenic Activity using Multi-Source Satellite-Based Remote Sensing Imagery, https://ntrs.nasa.gov/api/citations/20240008645/downloads/document.pdf

4. Center for Geospatial Intelligence \- GMU College of Science \- George Mason University, https://science.gmu.edu/academics/departments-units/geography-geoinformation-science/research/center-geospatial

5. A framework for intelligence analysis using spatio-temporal storytelling \- People, https://people.cs.vt.edu/\~clu/Publication/2015/Geoinformatica-2015.pdf

6. Configurable Spaces: Temporal Analysis in Diagrammatic Contexts \- Uncharted, https://uncharted.software/assets/configurable-spaces.pdf

7. An Interactive Web Based Spatio-Temporal Visualization System \- University of Southern California, https://sail.usc.edu/\~akramakr/pdfs/spatio-temporal-visualization.pdf

8. Clone Deterministic 3D Worlds with Geometrically-Regularized World Models \- arXiv, https://arxiv.org/html/2510.26782v1

9. Enhancing the Visibility of Labels in 3D Navigation Maps \- Department of Computer Science, https://www.cs.cit.tum.de/fileadmin/w00cfj/cg/Research/Publications/2012/Enhancing\_Visibility/2012\_Vaaraniemi\_Enhancing\_the\_Visibility\_of\_Labels\_in\_3D\_Navigation\_Maps.pdf

10. Stories in GeoTime \- Uncharted, https://uncharted.software/assets/stories-in-geotime.pdf

11. \[2206.09910\] Timeline Design Space for Immersive Exploration of Time-Varying Spatial 3D Data \- arXiv, https://arxiv.org/abs/2206.09910

12. WAI-ARIA Graphics Module Published | 2015 | Blog \- W3C, https://www.w3.org/blog/2015/12/wai-aria-graphics-module-published

13. How does the Shadow DOM appear in the accessibility tree? \- Max Design, https://www.maxdesign.com.au/articles/the-shadow-dom.html

14. Accessible Hardware Accelerated Graphics (and WebGL) \- Quorum Programming Language, https://quorumlanguage.com/tutorials/accessibility/accessibleGraphicsWebGL.html

15. Categorization of Affordances and Prediction of Future Object Interactions using Qualitative Spatial Relations, https://etheses.whiterose.ac.uk/id/eprint/33571/1/Alexia\_Toumpa\_PhD\_Thesis\_WhiteRose.pdf

16. Smartflow: Enabling Scalable Spatiotemporal Geospatial Research \- arXiv, https://arxiv.org/pdf/2506.03022

17. The Sandbox for Analysis – Concepts and Methods \- Uncharted Software, https://uncharted.software/assets/the-sandbox-for-analysis.pdf

18. Sensemaking with Shared Visualizations: Investigating the Effects of Visualizations in Remote Collaborative Analysis \- SCS TECHNICAL REPORT COLLECTION \- Carnegie Mellon University, http://reports-archive.adm.cs.cmu.edu/anon/hcii/CMU-HCII-11-104.pdf

19. Force-Directed Drawing Algorithms \- Brown CS, https://cs.brown.edu/people/rtamassi/gdhandbook/chapters/force-directed.pdf

20. Force Directed 3D Graph Visualization Algorithm \- Preprints.org, https://www.preprints.org/manuscript/202404.1504

21. Full article: Cartographic Design as Visual Storytelling: Synthesis and Review of Map-Based Narratives, Genres, and Tropes \- Taylor & Francis, https://www.tandfonline.com/doi/full/10.1080/00087041.2019.1633103

22. Immersive and interactive visualization of 3D spatio-temporal data using a space time hypercube: Application to cell division and morphogenesis analysis \- Frontiers, https://www.frontiersin.org/journals/bioinformatics/articles/10.3389/fbinf.2023.998991/full

23. Guidelines for the Effective Design of Spatio- Temporal Maps \- International Cartographic Association, https://icaci.org/files/documents/ICC\_proceedings/ICC2013/\_extendedAbstract/443\_proceeding.pdf

24. BinoForce: A Force-Based 3D Dynamic Label Layout Method Under Binocular Viewpoints, https://www.mdpi.com/2079-9292/14/11/2199

25. Labels on Levels: Labeling of Multi-Scale Multi-Instance and Crowded 3D Biological Environments \- PMC, https://pmc.ncbi.nlm.nih.gov/articles/PMC8620714/

26. Mixed Labeling: Integrating Internal and External Labels \- Algorithms and Complexity Group \- TU Wien, https://www.ac.tuwien.ac.at/files/pub/Cmolik2020.pdf

27. Optimized 3D Point Labeling with Leaders Using the Beams Displacement Method \- arXiv, https://arxiv.org/abs/2407.09552

28. 3D force-directed graph component using ThreeJS/WebGL \- GitHub, https://github.com/vasturiano/3d-force-graph

29. 3D Force-Directed Graph \- GitHub Pages, https://vasturiano.github.io/3d-force-graph/

30. spark: modular, composable shaders for graphics hardware a dissertation submitted to the department, http://graphics.stanford.edu/\~tfoley/papers/tfoley-dissertation.pdf

31. Label placement in 3D georeferenced and oriented digital photographs using GIS technology \- GDMC, https://gdmc.nl/publications/2007/Label\_placement\_digital\_photographs.pdf

32. Semantic Exploration of Text Documents with Multi-Faceted Metadata Employing Word Embeddings: The Patent Landscaping Use Case \- FIZ Karlsruhe, https://www.fiz-karlsruhe.de/sites/default/files/FIZ/Dokumente/Forschung/ISE/Masterarbeiten/Masterthesis-Tatyana-Skripnikova.pdf

33. Focus Plus Context Screens: Combining Display Technology with Visualization Techniques \- Patrick Baudisch, http://www.patrickbaudisch.com/resume/7.%20samplePublicationsPatrickBaudisch.pdf

34. Labeling Out-of-View Objects in Immersive Analytics to Support Situated Visual Searching \- Johanna Beyer, https://johanna-b.github.io/files/documents/labels\_lin\_2021.pdf

35. Automated accessibility scanning and the Shadow DOM \- Esri Community, https://community.esri.com/t5/accessibility-blog/automated-accessibility-scanning-and-the-shadow/ba-p/1698393

36. Full accessibility tree in Chrome DevTools | Blog, https://developer.chrome.com/blog/full-accessibility-tree

37. Accessibility tree \- Glossary \- MDN Web Docs \- Mozilla, https://developer.mozilla.org/en-US/docs/Glossary/Accessibility\_tree

38. Visualizing Graphs in 3D with WebGL | by Michael Hunger | Neo4j Developer Blog | Medium, https://medium.com/neo4j/visualizing-graphs-in-3d-with-webgl-9adaaff6fe43

39. WAI‑ARIA Guidance: Best Practices for Accessible Web Interfaces, https://www.levelaccess.com/blog/wai-aria-guidance-best-practices-for-accessible-web-interfaces/

40. Graphics Accessibility API Mappings \- W3C on GitHub, https://w3c.github.io/graphics-aam/

41. WAI-ARIA Graphics Module \- W3C, https://www.w3.org/TR/graphics-aria-1.0/

42. Mastering ARIA Accessibility for Modern Web Design, https://beaccessible.com/post/aria-accessibility/

43. What the Heck is ARIA? A Beginner's Guide to ARIA for Accessibility \- Lullabot, https://www.lullabot.com/articles/what-heck-aria-beginners-guide-aria-accessibility

44. WAI-ARIA basics \- Learn web development | MDN, https://developer.mozilla.org/en-US/docs/Learn\_web\_development/Core/Accessibility/WAI-ARIA\_basics

45. Track: Poster Session 6 \- NeurIPS 2026, https://neurips.cc/virtual/2022/session/64307