Rooke Poole builds evidence-aware AI systems, local-first software, machine languages, and computational models of reality.
Enter selected work ↓The practice
Build the strange thing.
Then make it answerable.
Research architecture
Local-first products
Native systems
Executable experiments
Selected systems / 06
Not a highlight reel. Six active lines of inquiry spanning reasoning, authority, computation, and the machine itself.
A constraint-driven reasoning and verification stack for research, audits, and exact-binary provenance—built to preserve hard requirements, challenge assumptions, and replay final invariants.
A WebMCP incident workbench where evidence, authority, mitigation, and audit state share one surface—and the consequential action does not exist until a human approves it.
A public computational-physics program exploring a three-dimensional cellular automaton, emergent dynamics, and falsifiable boundaries—with experiments, known gaps, and a status ledger.
A family of QML-native tools for Omarchy: system control, Codex packaging, focus and breathing, lunar phase, palette generation, decision support, telemetry, and a full Oligarchy shell experience.
A source-available UEFI-to-desktop operating-system effort centered on an original loader, capability-based microkernel, isolated services, PooleGlyph execution, and a PooleGlass desktop.
A collection of exact-byte executable experiments—including SpaceGlass—focused on direct reconstruction, runtime evidence, failure history, and honest provenance classification.
Public work index
Complete original-project index from the public GitHub profile as of August 27, 2026. This portfolio repository and twenty-nine forks are intentionally excluded.
About Rooke
Across the portfolio, one thesis repeats: authority should be explicit, local software should remain accountable to its operator, and extraordinary technical claims should carry evidence proportional to the claim.
The work ranges from desktop utilities and research infrastructure to experimental machine languages, operating systems, agent coordination, computational physics, and direct executable construction.
Keep the user close to their data, tools, and consequential decisions.
Separate what was observed, reproduced, inferred, and still needs independent validation.
Make speculative work inspectable, executable, and honest about where it breaks.