A compact origin.
Generate structural states from a defined seed and rule set. Matrix Controllers express that structure through a compact, regular interface.
Holo-Geometric Computing / Torus³ Platform
Separate layers.
One mathematical
structure.
A generative, nonlinear topological computing fabric that unifies addressing, routing, scale, recovery, and verification through a shared control grammar.
NESTED TORUS³ DATAFIELD / STRUCTURAL VIEW
DRAG TO ROTATE
One field. Multiple contexts.
The Matrix Controller is the linear interface to a nonlinear datafield. Plane, scale, and orientation specify how a structural location is interpreted.
01 / What is being unified
The structure does
more than hold
the computation.
In Torus³, the mathematical relationships that generate the field also provide the basis for navigating it, changing resolution, checking consistency, and recovering supported structural states.
That is the central architectural idea: functions often managed through separate descriptions are expressed as connected operations and observations of one algebraic structure.
Computational functions
AddressRouteScaleRecoverVerify
One mathematical structureMatrix Controllers · Universal Scale Kernels
SHARED GENERATING RULESCONNECTED STRUCTURAL CHECKS
The same field is read and operated through different perspectives. A new view does not require a new, unrelated architecture.
A connected control grammar.
Offsets, orientations, planes, scales, and Scale Kernels change how the same structure is selected, related, and traversed.
Verification is connected.
Symmetry, recurrence, and conservation offer mutually related observations. Their agreement can support verification and qualified structural recovery.
02 / Understand the platform
Two perspectives.
The same foundation.
These are the mechanisms that make the unification concrete. Each feature describes an operation or relationship within the Torus³ structure, with a link to the scope of its programmatic evidence.
✓ Programmatically verified
A passing recorded test supports the stated scope. Expand a check for coverage; it is not a blanket performance certification.
Seed-generated topology
A compact generating structure produces the native Matrix Controllers and their structural relationships. The field is defined by its rules, rather than by a separately specified value at every location.
✓ Programmatically verified
Native generation matched all 5,184 controller cells in the declared reference frame. Recurrence tests covered every coefficient seed in the finite domain.
VAL-REC-001 · VAL-CTR-001Closed controller families
Scalar, operator, and anchor controllers belong to one defined algebraic system. Their relationships supply a shared language for selecting, transforming, and checking structural states.
✓ Programmatically verified
Family actions, arithmetic role partitions, operator projections, and anchor supports passed their declared finite tests. This check does not certify a general scheduler.
VAL-FAM-001 · VAL-OPR-001 · VAL-ANC-001Nonlinear, plane-aware addressing
A carrier position can participate in multiple labeled address contexts. Keeping the plane and field context makes those addresses distinguishable and supports overlapping use of the same carrier.
✓ Programmatically verified
Every one of 13,824 labeled addresses was checked in a fixed native field, with structural lookups across nine controllers. Logical address contexts do not multiply physical payload storage.
VAL-PER-002Scale covariance & conservation
Coarse and fine observations belong to the same generated field. Correctly registered, complete partitions conserve whole totals; changing scale does not require an unrelated structural description.
✓ Programmatically verified
All nine controllers and eight native scales were covered. Conservation concerns whole-integer totals over matched supports; reduced-root arrays are not uniformly semi-magic at every scale.
VAL-SCL-001 · VAL-SMG-002 · VAL-AGG-001 · VAL-TV-001Universal Scale Kernel connectivity
A finite library of Universal Scale Kernels supplies the relationships used to connect supported scales. Refinement retains lineage and phase so a finer view stays attached to its structural context.
✓ Programmatically verified
Target generation, parent return, and supported refinement paths passed. The refinement audit covered 3,132 aligned child blocks across 12 proper divisible kernel-size pairs.
VAL-KER-001 · VAL-KER-002 · VAL-KER-003Offsets & the Dihedral Controller
Wrapped offsets, rotations, and reflections provide composable ways to act on the field. Orientation and registration travel with the operation so a transformed view remains interpretable.
✓ Programmatically verified
Offset closure was checked across native scales. Eight dihedral actions and their 64 ordered compositions were tested at each supported square scale, alongside kernel covariance.
VAL-OFF-001 · VAL-D4-001 · VAL-KER-005Composed, scale-aware routing
Routes combine structural operations with plane-aware destinations. Supported scale changes retain both the coarse selection and local phase, allowing the tested transformations to be reversed.
✓ Programmatically verified
The native-plane campaign covered 344 elementary configurations at every labeled address and 64 seeded eight-step compositions. It does not cover every possible route string or a complete packet scheduler.
VAL-ROU-002 · VAL-ROU-004Controller-Braid expansion
Larger fields interleave complete native controller fibers through ordered Scale Kernel expansion. A selected fiber is distributed across the larger field while retaining its own complete controller structure.
✓ Programmatically verified
Seven ordered expansion grammars were tested across nine controllers, including 1,377 complete fibers. Axis shifts were exhaustive in the tested fields; two-axis compositions were sampled.
VAL-BRD-001 · VAL-BRD-002Immutable payload binding
The structural address, its payload reference, and the stored payload are distinct. A lawful binding transformation can change how data is addressed while leaving the stored bytes unchanged.
✓ Programmatically verified
13,824 distinct synthetic payloads were tested across a full plane cycle, isolated updates, collision controls, and independent exported-snapshot decoding. Payload bytes were retained in a separate store.
VAL-PAY-001Antipodal & transpose symmetry
Complementary positions and transpose relationships connect different parts of the same field. These relationships provide additional ways to interpret and cross-check structural observations.
✓ Programmatically verified
Native half-turn and transpose relations passed. Scale-level value complements and legal spatial half-turns were tested separately; odd scales do not acquire an integer half-side shift.
VAL-ENT-003 · VAL-ENT-004 · VAL-SYM-001Self-healing structural recovery
Surviving information, structural relationships, and trusted context can support restoration of missing structure. Recovery and deliberate regeneration are separately tracked operations.
✓ Programmatically verified
Declared single-seat, contiguous, seeded multi-erasure, and mixed-fault cases were checked, including required refusals. The separate physical FPGA campaign is summarized below.
VAL-RECOV-001 · VAL-RECOV-002Consensus-based verification
Recurrence, symmetry, and conservation supply connected consistency checks. Agreement among retained structural observations supports recovery; conflicting or insufficient evidence can require refusal.
✓ Programmatically verified
Conservation and symmetry checks, contradictory-witness controls, and ambiguity refusals passed in their recorded domains. Consensus here concerns structural observations, not an implemented distributed consensus protocol.
VAL-SMG-001 · VAL-TV-001 · VAL-RECOV-001 · VAL-RECOV-002Algebraic closure: the Formal Algebraic System v1.0 has 88 project-accepted written derivations within its fixed-reference scope. Mathematical closure and finite programmatic test coverage are distinct evidence layers.
The architectural thesis
Reduce the work needed
to organize the work.
Computing systems need more than arithmetic. They also need ways to locate information, relate levels of detail, maintain mappings, route requests, and determine whether results remain consistent.
Torus³ brings these structural functions into a common mathematical framework. If the relationships can be generated and reused efficiently, some of the tables, repeated transformations, and coordination work surrounding useful computation may be reduced.
The opportunity lies in turning this unification into measurable implementation value: fewer structural memory accesses, compact control logic, and reliable operation across connected views.
One foundation across multiple functions
A shared controller and Scale Kernel framework connects generation, addressing, scale, and verification. The differentiation is the integrated construction and its defined relationships.
✓ Underlying mechanisms tested in declared domainsData can stay in place as views change
Tested payload bindings preserve stored bytes while structural references transform. This creates a concrete basis for investigating alternatives to repeated data rearrangement.
✓ Immutable payload binding verifiedAn engineering program with physical evidence
Core operations have reached developmental FPGA demonstrations. Matched implementations are being compared to determine where generation helps and what it costs.
✓ Scoped physical recovery campaign matched reference88Project-accepted mathematical derivations
47Passing records in the retained validator suite
288Cases in the physical recovery campaign
The investment case is an architecture and an evidence-building program. Workload-level speed, energy, and resource advantages must come from matched measurements.
03 / One controller. Connected field views.
See the relationship
between the views.
The square Matrix Controller is a chart of a periodic field. Use its actual 24-position structure to select an aggregate or distributed phase, then see the same support highlighted in the datafield.
Embedded Matrix Controller / 24 × 24
Click a location, or use the arrow keys. The chart and datafield show the same selected support.
CONTIGUOUS AGGREGATE SUPPORT16 positions
CONTROLLER AND FIELD / ONE ADDRESS SUPPORTDRAG TO INSPECT
One complete blockSelected position: (6, 9)
Aggregation reads a contiguous block of the native field. The whole-integer total over a complete partition is conserved; the total alone does not encode the arrangement inside the block.
Geometry follows the supplied Ratio Navigator and Datafield Lab. Coordinate supports are interactive; protected structural values, generating sequences, kernel coefficients, and recovery logic are omitted. Coincident projected points remain distinct logical addresses.
04 / Why this matters to computing
Faster arithmetic
doesn’t remove
the cost of access.
In a conventional processor–memory system, computation depends on moving instructions and data between separate resources. Waiting for those transfers can limit useful work. This is the core of the von Neumann bottleneck.
Torus³ approaches a related part of the problem: how much structural information must be stored, fetched, rearranged, and coordinated before the useful work can happen?
Reduce structural lookups.
When an address relationship or control state can be derived locally, an implementation may replace a structural table read with logic. That could reduce pressure on memory resources used to describe the structure.
Avoid unnecessary rearrangement.
Some operations can be expressed as changes to selectors, frames, or bindings. Keeping payload bytes fixed while changing how they are addressed can avoid moving data solely to construct another view.
Connect control and checking.
Scale, routing, and structural checks share established relationships. Hardware can investigate reusing those relationships across functions, reducing the need to maintain disconnected representations of the same structure.
What changes in the engineering question?
Structural memoryStored mapping or control value
OperationUse the requested relationship
A table-based design retrieves the stored structural value it needs. The engineering cost includes that access and the storage used to hold the description.
Architectural comparison only. No transfer counts, timing, or energy measurements are simulated.
The practical goal is to relieve selected sources of structural overhead. Payload access still has a cost. Torus³ does not yet establish that the von Neumann bottleneck is solved; matched FPGA workloads are the next step in measuring how much this approach changes the balance.
Background on the processor–memory data-movement problem: Kingra et al., SLIM (2020). This background source explains the general bottleneck; it is not evidence for Torus³ performance.
05 / Expand without losing the relationship
A larger field.
A complete controller
within it.
Controller-Braid expansion distributes complete native controller fibers across an expanded datafield. Each selected phase picks out a whole controller at a regular spacing.
The larger structure is an interleaving, with the controller relationships carried through the expansion. The expanded view does not simply restart an isolated controller at every adjacent tile.
Change the field size, then move through its phase fibers. The green fiber retains 576 controller positions as its locations spread through the expanded field.
✓ Complete fibers verified in seven tested expansion grammarsFIELD SIDE 48
FIBER SPACING 2SELECTED FIBER 576 positions
PHASE FIBERS 4
FIBER SPACING 2SELECTED FIBER 576 positions
PHASE FIBERS 4
Fiber membership and spacing follow the supplied interleaving model. The gray background is visually sampled. Controller-family values and kernel generation are not exposed.
06 / From formal structure to physical behavior
Implemented.
Observed.
Checked.
Torus³ has moved through a consolidated mathematical framework and repeatable software tests into developmental hardware demonstrations on the Digilent Arty A7-100T. The physical recovery campaign provides a concrete checkpoint.
264,600
Composed-operation observations
Stage 6 physical FPGA capture288
Physical test cases
Recovery, regeneration, controls, refusals0
Recorded mismatches
Against the campaign referenceSeptember 18, 2026 · Defined structural states, trusted context, and declared erasures in one native field at a time. This result includes correctly refusing unsupported recoveries.
198Survivor recovery cases
72Expected refusals
9Explicit regenerations
9Healthy controls
MATHEMATICS
A closed, declared foundation.
88 project-accepted written derivations in Formal Algebraic System v1.0 establish results within the fixed-reference mathematical scope.
PROGRAMMATIC VALIDATION
Coverage attached to claims.
The retained Validator Suite v1.1 records 47 passes and two tests not run. Feature checks above link to relevant passing records and their limits.
HARDWARE DEVELOPMENT
Comparative measurement next.
Generated, table-based, and hybrid FPGA implementations have completed reported build checks. Reviewed build summaries do not yet include their physical benchmark outputs.
Evidence definitions and current boundaries
Programmatic checks refer to recorded tests, not new suite runs performed by this page. Some test domains are exhaustive; composed sequences and expanded examples have explicitly bounded coverage. The general typed-route contract and general regional-decoding/scheduling gate were not run in the retained suite.
The physical campaign reports 45 insufficient-witness refusals and 27 conflict refusals. Structural recovery, deliberate regeneration, arbitrary application-data recovery, and performance measurement are different claims. The first two have scoped evidence here; general payload recovery, energy savings, and a workload-level performance advantage are not established by these records.
Sources reviewed: Formal Algebraic System v1.0; Master Architecture Manual v1.0; Validator Master Report v1.1; Stage 6 physical capture; and the later comparative build summaries. The geometry is adapted from the supplied Tensor Controller Ratio Navigator v12 and Datafield Lab v0.3. Protected source reports and algorithms are not bundled with the page.
07 / Where the architecture could matter
One foundation.
Several engineering directions.
Applications follow from what the structure can support. Each direction requires its own workload, implementation, and evidence.
Routing fabrics
Investigate compact structural addressing and composed routes for reconfigurable interconnects and network-on-chip research.
Edge hardware
Evaluate generation versus stored structural tables when memory resources, logic area, and predictable behavior matter.
Multiscale workloads
Explore coherent coarse-to-fine views for spatial systems, XR, and media workflows, with payload codecs evaluated separately.
AI memory organization
Investigate explicit relationships and address contexts for organizing retrieval. Model quality and application benefits need their own measurements.
The opportunity starts
with unification.
Torus³ connects the structure that generates, addresses, routes, and checks. The engineering work turns those relationships into measured value.
Development overview · September 2026. Green checks identify the recorded scope of programmatic verification. Protected controller values and implementation algorithms are omitted.