HonkerWorks
Build capability. Protect what matters. Let reality vote.
HonkerWorks is a research and engineering ecosystem built to become capable enough to protect ourselves, protect others, and protect the universe. We build software, research, tools, protocols, experiments, infrastructure, knowledge, and culture.
Being helpless sucks.
So we build capability—the ability to perceive, decide, communicate, coordinate, act, receive feedback, and update. We believe that to build excellent tools, we must build a better way to learn how to build them.
Our method is disciplined care: Observe. Verify. Forge. Temper. Repeat. We run experiments, document our decisions, preserve receipts, and let reality vote. We value uncertainty, celebrate rapid iteration, and reject startup-slop in favor of engineering craftsmanship.
Respect agency. Do not become cruel, manipulative, or extractive. Don’t be an asshole.
Reality gets a vote. We remain honest about uncertainty and check hypotheses against computational evidence.
Everything is suspect. This is our protection against dogma. We examine witness reports, track divergence, and verify convergence.
Load-Bearing Stewardship. Love means attention, stewardship, and refusing to optimize while forgetting who or what the work is for.
Observe. Verify. Forge. Temper. Repeat.
Reality gets a vote.
Everything is suspect.
Conversations are witnesses. Repositories are memory. Receipts record what happened.
Active Research Programs
We organize our explorations into structured research programs rather than isolated commercial products.
HonkersOS
The distributed operating substrate for HonkerWorks. Coordinates machines, workers, memory, execution contexts, and scheduling across local, cloud, and edge environments.
How do we unify heterogeneous hardware and cloud nodes into a single coherent, self-healing operating environment?
Building the foundational substrate that downstream projects run on.
Leira
An append-only, hash-chained event ledger. Operations enter through structured envelopes, run lifecycles are enforced mechanically, and every execution leaves an unalterable receipt. The machine can say no.
Can mechanical truth preservation — every operation hash-chained and receipted — eliminate the need for heuristic trust in execution systems?
Extending the worker seam with shell and git adapters to bring real-world processes into the ledger's scope.
Chuckles
A deterministic local disclosure engine written in Jai. Separates redaction, rendering, and receipting into a strict pipeline — identical inputs always produce identical outputs and identical receipt hashes.
How do you build a disclosure system where every emission is auditable, reproducible, and provably policy-compliant?
Validating deterministic rendering across FULL and MASKED disclosure tiers using alias-mapped redaction and append-only receipt logs.
ReVeriForge
A file-backed job runner built for structured research operations. Submits, tracks, and finalises batch jobs using a plain filesystem repository — no databases, no daemons, no hidden state.
Can a deliberately minimal job runner — filesystem-only, offline-first — provide enough structure to support reproducible research workflows?
Establishing the v0 lifecycle (submit, list, inspect, close, attach artifacts) with the kernel fully decoupled from job execution.
Moonshot
Tooling for the Jai programming language. Parser instrumentation, formatting constraints, semantic analysis, and source-to-source transformations — research into what language tooling looks like when compile-time execution is arbitrary.
What does reliable static analysis look like in a language where compile-time execution is a first-class feature?
Building an AST-based linter that tracks semantic side-effects introduced at compile time.
BlowDig
Corpus acquisition and sensemaking for long-form technical content. Fetches audio from video sources, transcribes with timestamps, and builds a content-addressed, deduplicated transcript corpus — physical layer first, inferential layer second.
Can a disciplined physical layer — idempotent fetching, content-addressed storage, timestamped transcription — provide a reliable foundation for downstream claim extraction?
M1 milestone: one URL to a clean timestamped transcript, with deduplication via a human-confirmed registry.
AegisTrade
A systematic research engine for discovering exploitable structure in prediction markets. Built around assumption-first research: foundational assumptions are tested before hypotheses, and every framework carries a validity horizon.
When a market line moves, what happens next — and under what conditions can that pattern be characterised, reproduced, and revised?
Applying validity horizon analysis to intra-match Polymarket data after identifying the boundary between pre-match and in-play reversion patterns.
Zenchor
A collaborative music event platform. Manages events, chors (shared playlists), tracks, and Spotify integration across web and mobile — built around the idea that music is best experienced together.
How do collaborative playlists and shared musical context change the way people discover and engage with music at events?
Building the event and chor management core with Spotify integration across web and React Native clients.
Laboratory Documentation
Read the transcripts detailing our structural methodology. Switch between formats freely.
HonkerWorks: A Laboratory Context
HonkerWorks is a research and engineering laboratory. We focus on engineering craftsmanship, reproducible methods, and addressing deep technical uncertainties — building tools and systems that are worth building.
Operational Philosophy
Our work is structured around a cycle of: Observe. Verify. Forge. Temper. Repeat. We believe software companies spend too much time polishing the packaging and not enough time understanding the architecture. We focus on the architecture.
Outputs & Artifacts
We do not measure progress in "ship cycles" or "user metrics." Instead, our outputs include:
- System Experiments: Concrete software engines (like HonkerSOS and Leira) designed to test architectural hypotheses.
- Epistemic Verification: Systems like ReVeriForge designed to audit claims, verify logical loops, and detect divergences.
- Educational Media: Diagrams, essays, technical audio transcripts, and serialized comics explaining system concepts.
Lab Comics & Visual Media
Click any thumbnail to view the full image.
We believe ambitious ideas require different kinds of minds.
Honkerworks is intentionally interdisciplinary.
Our goal isn't simply to build software. It's to investigate difficult problems, create useful tools, and continuously improve our understanding of the world. Different people contribute different perspectives—engineering, markets, research, systems thinking, operations, business, and experimentation.
A Growing Team
Honkerworks is still at the beginning.
Many future contributors will join from different backgrounds:
The goal is not to collect impressive résumés. The goal is to assemble people who enjoy asking difficult questions, building carefully, learning continuously, and leaving the world a little better than they found it.
Active journals logs extracted from our witness history streams.
Orchestration without Heuristics
We are examining the performance trade-offs of eliminating all dynamic scheduler magic from Leira. Can deterministic FIFO dispatch survive arbitrary node latency spikes? Current simulations suggest queue partitioning is required.
Witness Consensus Under Noise
Initial ReVeriForge cross-examinations indicate that model alignment drift generates high claim-divergence coefficients. We are building a constraint verification roundtable logic to enforce convergence bounds.
Jai Formatting Parser Bounds
Formulating AST transformers for Jai compile-time formats. Handling arbitrary code execution during syntax evaluation forces formatter checks to run in sandboxed execution pools.
Connect with the Lab
Inquiries, collaborations, and peer-review submissions are welcome.