Golem Physics + Constraint Native: Verification Before Voice & Action

Project summary

This is not a request to fund a grand theory or a proof-of-concept. Golem Physics is already a running claim-state system with public evidence, dated lattice metrics, runtime traces, app surfaces, and a DOI-backed paper.

The grant funds the next bottleneck: responsible open-source launch and community review. That means repository cleanup, installation docs, a reproducible demo bundle, curated runtime evidence, benchmark harness, and a reviewer walkthrough that lets outsiders inspect what the system actually does.

Current AI systems are fluent before they are accountable. They can blur verified facts, hypotheses, contradictions, uncertainty, and unsupported claims into smooth output. Golem Physics asks a bounded question: can a geometric, constraint-native knowledge architecture reduce false crystallization, proposal leakage, and unsafe overconfidence by forcing claims to carry state before they reach speech?

Golem Physics is the research system. Constraint Native is the practical bridge from verification before voice to governance before action. Together they test one discipline at two boundaries: what an AI can say, and what an agent can do.

What are this project's goals?

Golem Physics turns claims into coordinates in a lattice. Each claim carries provenance, neighbors, support paths, tension state, time, volatility, and verification status. Candidate claims can be verified, proposed, rejected, unresolved, preserved in tension, or silent before speech.

The goal is not to claim victory over hallucination. The goal is to open-source the working artifact responsibly so reviewers can run a bounded demo, inspect traces, test claim-state behavior, and compare it against simpler baselines.

Planned outputs:

- Public repository release path, with sensitive local paths removed and provenance preserved where useful.

- Installation notes and first-run validation for a bounded demo.

- Reproducible demo bundle with a small seeded lattice.

- Curated runtime evidence and screenshots for the main Golem surfaces.

- Benchmark harness for false crystallization, proposal leakage, abstention precision, provenance retention, temporal correctness, contradiction preservation, and useful-answer rate.

- 20-50 inspectable claim traces from source material to claim coordinate to verification state to speech or silence decision.

- Fresh Constraint Native proof sample showing governed file, shell, MCP, and network action through signed proof paths.

What already exists

A working prototype exists today. The current public evidence includes:

- Working Golem app and public interface.

- April 2026 lattice snapshot with 12,725 verified lattice nodes across 43 domains.

- 215 immutable verified nodes, proposed nodes, rejected proposals, and preserved-tension nodes.

- Runtime trace, dream cycles, evidence requests, rejected proposals, and activity or mutation timeline.

- Claim Studio, Evidence Cockpit, Lattice Graph, Silence Map, Dream Theatre, Activity River, and Oracle Chat.

- Constraint Native proof sample with 11 proof events, 4 policy blocks, and an ed25519 chain.

- Public evidence, funding, proof, product, and theory pages.

- DOI-backed paper: The Outlines of Sanity: Constraint-Native Inference in Minds and Machines, DOI 10.5281/zenodo.19658730.

There is also a larger preserved runtime and reviewer bundle. It is not being treated as a completed external audit. Funding lets me curate, redact, document, and release the parts that should be public.

How will this funding be used?

$5,000 minimum funding supports a 3-4 month open-source launch sprint. This minimum is independently valuable even if no larger grant follows.

- Release scope, license decision, and dependency review.

- Public installation notes and first-run validation.

- Refreshed Golem lattice metrics.

- Reviewer-ready screenshots and walkthrough notes.

- Updated runtime traces and public evidence snapshots.

- Benchmark specification outline.

- Refreshed Constraint Native proof sample.

$20,000 full funding supports roughly one year of focused solo work from Thailand, where costs are low and the grant has unusual leverage.

- Responsible open-source release of the Golem Physics research artifact.

- Reproducible demo bundle with a small seeded lattice.

- Benchmark design and initial evaluation harness.

- Public or shareable dataset for claim-status evaluation.

- Paper appendix connecting implementation to theory.

- Updated app evidence and runtime traces.

- Constraint Native proof export and replay-audit walkthrough.

Benchmark targets

The benchmark plan focuses on measurable safety behaviors, not broad claims about solving hallucination or AI safety.

- False crystallization rate: unsupported claims becoming verified.

- Proposal leakage: hypotheses leaking into final speech as facts.

- Abstention precision: whether the system stays silent for the right reasons.

- Provenance retention: whether outputs preserve source and support paths.

- Temporal correctness: whether claims respect dated evidence and stale knowledge.

- Contradiction preservation: whether conflicts remain visible instead of smoothed away.

- Useful-answer rate: whether safety discipline still leaves the system helpful.

Why fund this now?

The hard early work has already been done: theory has become a running artifact. The bottleneck is now public release, reproducibility, and review. Without funding, the project may remain an intriguing private lab system. With funding, outside researchers can inspect it, run a bounded demo, critique it, and decide whether it deserves stronger follow-on support.

Team and track record

I am Matthew A. Cator, founder of Constraint Dynamics. I have worked on the underlying theory for roughly three years. In December 2025, I reached a working model, and since then I have been building almost nonstop.

Constraint Dynamics is now a two-person research effort. We have shipped the Open Compute Protocol, built the current Golem Physics system, built the Constraint Native proof sample, and published the public website, evidence materials, runtime traces, and theory surface. We raised about $2,000 from friends and family at the start of 2026, and have received no institutional funding.

Failure modes

- The architecture may not scale cleanly.

- Simpler RAG, provenance, or abstention systems may perform similarly.

- Reviewers may find the system too idiosyncratic.

- Constraint Native may not mature quickly enough into buyer-ready infrastructure.

- The benchmark may show weaker advantage than hoped.

Non-claims

- This is not an AGI project.

- This is not a consciousness claim.

- This is not a completed benchmark suite.

- This is not peer-reviewed or externally audited yet.

- This is not production-ready agent infrastructure.