# SOKR — References
> Curated references that inform SOKR's design decisions.
> Organized by relevance to core architectural concerns.
---
## Formal Foundations
### Automata & State Machines
- **Automata Theory** — formal basis for computation models that don't assume
a specific execution substrate
- **Statecharts** (David Harel, 1987) — hierarchical state machines with
well-defined semantics for concurrent and reactive systems
- **RTC — Run-to-Completion semantics** — execution model where a state
machine processes one event completely before accepting the next.
Relevant to SOKR's Completion signal contract.
### Formal Verification
- **seL4** — https://sel4.systems
The only OS microkernel with a machine-checked formal proof of
correctness and security enforcement. Capability-based design.
The gold standard reference for what a sovereignty claim should
aspire to — not just philosophical, but mathematically provable.
Relevant to SOKR's long-term formal verification roadmap.
---
## Capability-Based Design
- **Capsicum** — https://www.cl.cam.ac.uk/research/security/capsicum
Capability-based security model for UNIX. Fine-grained sandboxing
via capability tokens. Informs SOKR's plugin isolation model.
- **cap-std** (Rust) — https://github.com/bytecodealliance/cap-std
Capability-based filesystem and network access for Rust. Early
concept predates WASI 0.2 complexity. Clean reference for
capability design in Rust.
- **WASI 0.1** — original WebAssembly System Interface runtime concept.
Clean capability-based design before committee consensus introduced
complexity. The simplicity of 0.1 is a cautionary reference for
SOKR — keep the core interface minimal.
- **WASI 0.2 / WIT / IDL** — https://github.com/WebAssembly/wasi
Where WASI introduced the WebAssembly Interface Type (WIT) IDL.
Cautionary reference: SOKR deliberately avoids IDL, schema
generation, and code generation at the core layer for this reason.
---
## Existing Runtimes
### General Purpose
- **BEAM** (Erlang/Elixir runtime) — https://www.erlang.org
Actor model runtime that is substrate-agnostic by design. Processes
communicate via message passing with no shared memory. Different
layer from SOKR but same principle: the computation model is
independent of the execution substrate.
- **QNX** — https://blackberry.qnx.com
POSIX-compatible microkernel OS with capability-limited message
passing. Everything communicates via explicit IPC. Core is minimal,
capabilities are declared. Closest OS-level analog to SOKR's design
philosophy.
### WebAssembly
- **WASI** — https://wasi.dev
WebAssembly System Interface. Runtime abstraction for WASM modules.
SOKR's WebGPU substrate plugin will interact with WASM targets via
this interface.
---
## Hardware Dispatch Models
### Minimal Dispatch Reference
- **RP2040 PIO** (Raspberry Pi Programmable I/O) —
https://datasheets.raspberrypi.com/rp2040/rp2040-datasheet.pdf
Eight independent state machines programmable without touching the
core processor. Each PIO program is a small instruction sequence
dispatched to a state machine. Closest hardware analog to SOKR's
three-function dispatch model — capability (does this SM support
this program?), dispatch (load and run), completion (signal done).
Relevant for SOKR's embedded/MCU substrate story.
---
## Compute Runtimes (Existing Landscape)
### GPU
- **CubeCL** — https://github.com/tracel-ai/cubecl
Multi-platform high-performance Rust compute language with JIT
compiler, automatic vectorization, and autotuning. Supports CUDA,
WebGPU, HIP/ROCm, Metal, SPIR-V. Built around the GPU
thread/workgroup model. Most mature Rust GPU compute project.
SOKR is designed to sit below CubeCL as a substrate plugin layer.
- **wgpu** — https://github.com/gfx-rs/wgpu
Cross-platform GPU abstraction in Rust over Vulkan, Metal, DX12,
WebGPU. Targets the GPU thread/workgroup model. SOKR's Vulkan
substrate plugin wraps `wgpu` or `ash` internally.
- **rust-cuda / rust-gpu** — https://github.com/Rust-GPU/rust-cuda
Rust compiler backends for NVIDIA CUDA (via NVVM IR) and SPIR-V
(via Vulkan). The Chimera demo proves write-once, run-anywhere GPU
kernels in Rust. SOKR's CUDA and SPIR-V IR plugins build on this
ecosystem.
- **ZLUDA** — https://github.com/vosen/ZLUDA
Drop-in CUDA replacement for non-NVIDIA GPUs via PTX → HIP
translation. Attacks CUDA vendor lock-in at the compatibility layer.
Complementary to SOKR which attacks it at the contract layer.
### Quantum
- **CUDA-Q** (NVIDIA) — https://developer.nvidia.com/cuda-q
Open-source QPU-agnostic platform for hybrid classical-quantum
workloads. GPU + CPU + QPU from a single program. Built on CUDA,
so vendor-locked to NVIDIA stack. SOKR's QPU substrate plugin
targets the same problem space without the vendor dependency.
---
## Neuromorphic Hardware
- **Intel Loihi / Loihi 2** — https://www.intel.com/content/www/us/en/research/neuromorphic-computing.html
Intel's neuromorphic research chip. Programmed via the LAVA
framework. SOKR's `sokr-neuro` plugin targets this via LAVA bridge.
- **SpiNNaker** (University of Manchester) —
https://www.manchester.ac.uk/research/groups/advanced-processor-technologies/projects/spinnaker
Massively parallel neuromorphic architecture designed to simulate
biological neural networks in real time. Event-driven, spike-based.
- **IBM TrueNorth** —
https://research.ibm.com/publications/truenorth-design-and-tool-flow-of-a-65-mw-1-million-neuron-programmable-neurosynaptic-chip
IBM's 1-million neuron neuromorphic chip. 65mW power consumption.
Spike-based computation with no shared clock.
- **NIR — Neuromorphic Intermediate Representation** —
https://github.com/neuromorphs/NIR
A common IR for neuromorphic computations across 11 platforms.
Python-based. Informs SOKR's `sokr-neuro` IR plugin design.
Published in Nature Communications (2024).
---
## Photonic Compute
- **Lightmatter** — https://lightmatter.co
Photonic computing hardware using light instead of electrons for
matrix operations. Target for SOKR's future `sokr-photon` substrate
plugin pending public SDK availability.
---
## Standards & Specifications
- **SPIR-V** — https://www.khronos.org/spir
Khronos Group binary IR for GPU shaders and compute kernels.
Vendor-neutral. Supported by Vulkan, OpenCL, OpenGL.
SOKR's primary IR for GPU substrate plugins.
- **OpenQASM 3** — https://openqasm.com
Open quantum assembly language. Standard IR for quantum circuits.
SOKR's IR for QPU substrate plugins.
- **PTX** (Parallel Thread Execution) —
https://docs.nvidia.com/cuda/parallel-thread-execution
NVIDIA's virtual ISA for CUDA. SOKR's IR for CUDA substrate plugin.
---
## Community Feedback
References surfaced through community discussion that shaped
SOKR's design decisions:
- QNX / capability + message passing analogy → confirmed SOKR's
three-function interface design
- WASI 0.1 vs 0.2 → confirmed no IDL in SOKR core
- seL4 → formal verification as long-term sovereignty proof target
- RP2040 PIO → minimal dispatch model as hardware analog
- Rabbit hole warning → confirmed scope constraint: core stays at
three functions, all complexity delegated to plugins
---
*Copyright 2026 The SOKR Project — MIT OR Apache-2.0*