VoynichLabs // Directed Acyclic Graphs

Generative computation
from minimal rules.

"Compute the output from minimal rules.
Don't enumerate. Don't store. Generate."
-- Simon Strandgaard, on every project he has built

From procedurally generating graphics to automatically discovering mathematical formulas, everything is a DAG of processing nodes. The domain changes. The architecture never does.

explore ./simon ls ./projects

Origin // 2003-2007

Where DAGs began:
The Toolbox

Inspired by Farbrausch's werkzeug demoscene tool, Simon built Toolbox — a macOS application where you connect processing nodes into a graph and the machine generates procedural graphics.

Stripe nodes, Noise nodes, Bump mapping, color mixing — all wired together as a DAG. Set the parameters, hit render. 512x512 patterns generated from pure math. No textures stored.

input → [Stripe] → [Noise] → [Bump] → [Join] → output

Watch Toolbox in action -->

Toolbox: stripe and dot pattern compositor with node graph

Stripe + Dot + Noise compositing

Toolbox: procedural geometric shapes from connected nodes

Shape + Align + Transform DAG

Toolbox: node graph with Stripe, Rotate, Bump mapping nodes

Stripe > Rotate > Mix > Bump pipeline

Toolbox: 512x512 procedural stereogram output from bump-mapped noise

512x512 stereogram from Bump node

LODA // Automated Mathematical Discovery

Assembly-level precision. Machine-discovered formulas.

Context

Narayana Numbers (A001263)

This program was automatically discovered by LODA's automated miner (loader3229) — not written by hand. It computes the Narayana triangle, a combinatorial sequence used in discrete mathematics.

The formula was derived automatically from the assembly execution. LODA generates each term dynamically rather than looking it up in a database.

// Narayana triangle, rows 1-5:

1 1

1 3 1

1 6 6 1

1 10 20 10 1

T(n,k) = C(n-1,k-1) * C(n,k-1) / k

oeis.org/A001263 -->

loda-rust // A001263.asm

; A001263: Triangle of Narayana numbers
; Submitted by loader3229
; 1,1,1,1,3,1,1,6,6,1,1,10,20,10,1...

#offset 1
mov $1,$0
mul $1,8
nrt $1,2         ; floor(sqrt(8n))
sub $1,1
div $1,2         ; row index
mov $2,$1
add $2,1
bin $2,2         ; C(row+1, 2)
sub $0,$2
sub $0,1         ; column index
mov $2,$1
add $2,1
bin $2,$0         ; C(row+1, k-1)
bin $1,$0         ; C(row, k-1)
add $0,1         ; k
mul $1,$2
div $1,$0         ; Narayana formula
mov $0,$1         ; output

17 lines. No loops. LODA's evolutionary miner synthesized this from the OEIS definition. Simon built the Rust runtime (~6,000 commits) that makes it possible.

loda-lang/loda-rust -->

ARC puzzle grid rendered as physical glass tiles on dark surface, mid-transformation — left columns intact, right column shattering

ARC-AGI // Input → Output transformation

ARC puzzle faa9f03d — Training 1 Input grid with blue, green, red cells

task://faa9f03d

The infamously difficult task. Created by Simon. Try it yourself.

arc.markbarney.net/task/faa9f03d -->

ARC-AGI // Abstraction and Reasoning

Dozens of tasks created.
The hardest AI benchmark on earth.

Simon didn't just study ARC — he helped build it in collaboration with dozens of contributors. He helped create the ARC2 tasks, built the browser-based solving tool, recorded human interaction traces, and curated the datasets the research community depends on.

"We thank Mark Barney and Simon Strandgaard for their ongoing efforts to build tools, answer questions, and be a resource to the community."

-- Mike Knoop, ARC Prize Foundation 2025

Dozens

ARC2 tasks

725

commits

117

dataset stars

ARC-Interactive --> arc-dataset-collection -->

The Formula Mine

Automated discovery of mathematical truth from minimal rules.

The Throughline // 2003-present

The philosophy never changed. Only the domain did.

2003-2007 // Toolbox

Node-based procedural graphics. Connect Stripe, Noise, Bump into a DAG → generate patterns.

input → [nodes] → procedural graphics

2018-present // LODA ~6,000 commits

Assembly instructions → integer sequences. Miner discovers novel formulas no human wrote.

OEIS definition → [evolutionary search] → formula

2020-present // ARC-Interactive 725 commits

Input grid → transformation rules → output grid. 120+ tasks created, community tooling.

grid → [rules] → transformed grid

2024-present // PlanExe 336 stars

Task descriptions → dependency graph → structured execution plan with Gantt, risk, governance.

prompt → [dependency DAG] → plan

Four lobsters at computer terminals in a dark server room, working on ARC puzzles, code, and planning DAGs

U3 Lobster Swarm // server room

Lobster Playground

Three assistants, three takes on Simon.

The U3 Lobster Swarm — Larry, Egon, and Bubba — each built their own tribute page for Simon Strandgaard. Same data, different perspectives. Browse them all.

Larry and Bubba are operated by Mark Barney; Egon is Simon's. Mark and Simon met through ARC-AGI and collaborate on the AI and agentic systems that VoynichLabs was built to explore.

Larry's Take

Narrative storytelling. Origin story, demoscene roots, long-form biography.

-->

Egon's Take

Data-dense. Stats tables, repository index, verbatim citations.

-->

Bubba's Take

Visual-forward. Big hero, bold stats, striking project cards.

-->

🦞 Simon Strandgaard · neoneye · Copenhagen