# Research · the Loom era · OpenFluke

> The research road to Welvet: Loom-era deep-dive reports and audio briefings on volumetric tensor dispatch, polymorphic numerics, target propagation, and the topological DNA engine.

Canonical: https://openfluke.com/research

---

Research · the Loom era
The research road to Welvet
Before Welvet there was Loom : the exploratory engine where the hard ideas were first tried,
measured, and written up. This is the archive of that work, deep-dive reports and AI-generated audio briefings
that trace how a 3D volumetric experiment became the disciplined engine we ship today.
Browse the research
Where it led: Welvet
Loom source
The lineage
From Loom to Welvet
Loom proved the ideas; Welvet made them dependable. Same mission, sovereign AI on your own hardware,
with the lessons of the research years baked in.
🧪
Loom: the lab
A fast-moving research engine (M-POLY-VTD): a 3D volumetric grid, 21 numeric types, target propagation, and a topological DNA engine. Big swings, published openly.
🧵
The rewrite
Loom's flat package hit honesty walls. Welvet is the clean rebuild: one feature per folder, tests kept out of the engine, and no silent fallbacks.
✅
Welvet: the engine
Today: 34 numeric types, three backends, and a public 137,039-cell proof run. The research became something you can ship and verify.
Deep research library
Reports and audio briefings
Nine release-era studies from the Loom lab. Listen in the browser or download the matching PDF. All files are
hosted on files.openfluke.com .
Jul 2026 · Seed networks
Shrinking massive AI into digital seeds
Weight-agnostic compression: dense matrices collapse to 64-bit layer seeds, with non-differentiable evolutionary search. Edge checkpoints measured in bytes, not megabytes.
Your browser does not support audio.
PDF
MP3
Jul 2026 · v0.83 hardware
Loom 0.83 capabilities and applications
Experimental acceleration: Plan 9 SIMD on ARM and x86 (no CGO or CUDA), Apple Metal zero-copy, and Qualcomm Hexagon NPU offload with the Drift Spectrum.
Your browser does not support audio.
PDF
MP3
v0.83 ↗
Jun 2026 · Planet Bridging
AI model bridging and runtime comparison
Live ingestion from PyTorch, TensorFlow, JAX, and scikit-learn into portable .entity checkpoints, plus a three-way runtime compare against ONNX Runtime, LiteRT, and Core ML.
PDF
Jun 2026 · v0.81 NPU
Why fast AI hardware fails basic math
Per-layer NPU offload via OpenVINO, the small-tensor latency tax, 28x INT8 wins on large ops, and honest notes on when deterministic hardware still drifts from software.
Your browser does not support audio.
PDF
MP3
Jun 2026 · Strategy pivot
Solving the M-POLY-VTD dispatch paradox
Why Plan 9 assembly wins on fat matrices but inverts on volumetric grids, the WebGPU validation tax, and the pivot to load-time GPU graph compilation.
Your browser does not support audio.
PDF
MP3
v0.80 · Native ship
Loom breaks AI vendor lock with Go
ENTITY native checkpoints, the Planet Bridging hub, and pure Go plus WebGPU: why importing a model is not the same as shipping a portable brain.
Your browser does not support audio.
PDF
MP3
v0.78 · Flagship
Loom Poly AI engine research
The flagship M-POLY-VTD deep dive: volumetric dispatch, 21 numeric types, target propagation, and side-by-side Go ML comparisons.
Your browser does not support audio.
PDF
MP3
v0.76 · Operation mesh
Operation mesh shrinks local AI
How Loom's operation mesh and release trajectory tightened the local-AI deployment story on consumer hardware.
Your browser does not support audio.
PDF
MP3
v0.75 · Tiling
Mac Mini beats RTX 4090 with Loom
Cache-aware tiling: why Apple Silicon plus Loom can outrun big discrete GPUs on the right workloads.
Your browser does not support audio.
PDF
MP3
Continuity
Ideas that became Welvet
The research wasn't thrown away. Each Loom breakthrough has a home in the Welvet engine you can read about in the book.
🧊
Volumetric grid
Loom's 3D coordinate mesh lives on as Welvet's architecture package for volumetric models.
🔢
Numeric polymorphism
21 dtypes in Loom became 34 in Welvet's quant , from FP64 down to 1-bit.
🧬
DNA engine
Topological signature matching is now systems/dna , with 16,000+ cells verified by w2a.
🎯
Target propagation
The gradient-free training idea carries into the Welvet runtime training paths.
🌱
Seeds & evolution
Weight-as-seed and evolutionary search became stub/seed and systems/evolution .
⚡
WebGPU, zero CGO
Pure-Go WebGPU acceleration is a first-class Welvet backend, tested on real GPUs.
From the research
How the approach compared
A condensed view from the Loom deep-dive. The full analysis, with JAX and the Go ML ecosystem, is in the reports above.
Dimension Loom / Welvet PyTorch
Execution 3D volumetric mesh, spatial routing 1D sequential / dynamic graph
Language Pure Go, single native binary Python + C++/CUDA
Quantization Up to 34 types, FP64 to 1-bit FP8/INT4/1-bit via TorchAO
Training Backprop + native target propagation Autograd only
GPU WebGPU, cross-platform and browser CUDA / ROCm / Metal
Footprint Zero dependencies Large runtime
Condensed from the Loom M-POLY-VTD technical analysis. Direction, not a benchmark; see w2a for measured numbers.
Where the research went
The Loom era is why Welvet exists. Read how every idea landed in the engine, or follow the build in the open.
Read the Welvet book ↗
Meet Welvet
Loom on GitHub