The Strategy

AI is moving too fast for journals. By the time a review cycle completes, we've shipped another architecture. The journal pipeline publishes archaeology; the open-access + open-source pipeline publishes the living system.

Everything goes open. Journals are the anchor, not the gate.

When we claim 0% hallucination on the provable path, a reviewer either believes us or doesn't. When the reasoning core is open and you can run it yourself — that's a different kind of validation.

Published Papers

All 13 papers are open-access preprints on Zenodo (full methodology, results, limitations, raw data). The canonical list lives at ORCID 0000-0002-3912-0080.

# Title Key Result DOI
101 The NuSy Brain Architecture 4-layer knowledge model, 11.3ms latency, 0% hallucination on the symbolic path DOI
102 Training Neurosymbolic Beings 95-100% provenance, 100-1000x faster than fine-tuning DOI
103 Sub-5ms Reasoning Performance P95: 3.25ms, 1,345x speedup, 0% hallucination DOI
104 Continuous Learning / Crystallization 97.6% precision, 0 LLM calls on learned knowledge DOI
107 Graph-Native Episodic Memory 85.7% mistake prevention, +33.3pp decision quality DOI
108 Unified Cognitive Architecture 9/9 hypotheses validated, 0% hallucination, 146x vs RAG DOI
110 Semantic Field of View 76.2% next-topic prediction, 42.8% auto-answer DOI
112 Hierarchical Concept Index 0.003ms lookup, O(1) at 1M concepts DOI
113 COG Transfer Experiment Equal quality, ~100,000x faster knowledge transfer DOI
114 Domain-Aware Hallucination Prevention 0 safety violations, 6 domains DOI
117/119/120 Cognitive Layer Architecture (Y0-Y6) Y5: 85.7-100% error elimination, Y6: 4.8% ECE DOI
118 Lean Hypothesis Testing 5/5 hypotheses failed — an honest-failure / methodology paper DOI
122 AGI Certification Framework 10-dimension ACF, Zorblaxia contamination-proof battery DOI

The preprints above cover the V7–V15 architecture. Newer campaigns: V18 (executable reasoning + the provable gate) and V19 (the NuSy LRM, below), with preprints in progress. V17 (causal vector-symbolic dedup) was tested and refuted — see below.

What didn't work — and why we publish it

Real science produces dead ends; hiding them is how hype accumulates. We version the graveyard, not just the wins:

This is the same discipline as the provable gate itself: abstain honestly rather than assert falsely.

V19 — the NuSy LRM: provable clinical reasoning

The current campaign is a Large Reasoning Model built on one hard rule: never let a guess masquerade as a fact. A neural layer proposes; a strictly separated symbolic layer proves — or abstains. On the provable path, hallucination is zero by construction. We're proving it first in the hardest auditable domain — clinical decision support (FHIR-CPG-grounded, UMLS/SNOMED-native) — because that's where a confident wrong answer is least acceptable. See the platform.

Current, honest, on-disk measurements from the V19 clinical batteries:

Measure First, Publish Second

The ACF (AGI Certification Framework) tells us whether provable cognition is improving — a feedback loop that runs on weeks, not publication cycles. It's one continuous pursuit: provable understanding, measured version over version.

The through-line. ACF began by measuring general provable cognition across education beings (below). V19's clinical reasoner is the next rung on that same ladder — the same never-launder discipline, now proven where correctness is non-negotiable. We believe provable clinical reasoning is a step toward general AGI, and the foundation for a safer one: the mechanism that makes it safe in medicine — causal (do-calculus) reasoning behind a symbolic proof gate, with a strong LLM confined to proposing — is domain-agnostic. Medicine is the evidence, not the ceiling.

V11 → V12 Benchmark (February 2026) — general provable cognition across four education beings (same corpus, same machine):

Being V11 V12 Delta Level Change
Toddler 47.9 56.6 +8.7 ACF-3 to ACF-3
Gradeschool 58.8 63.3 +4.5 ACF-3 to ACF-4
Middleschool 47.6 63.3 +15.8 ACF-3 to ACF-4
Highschool 49.2 63.3 +14.1 ACF-3 to ACF-4

Consistent improvement across every level; the biggest gains in compositional generalization and factual grounding — the same properties the V19 provable gate now enforces clinically.

How we work

Every change carries a falsifiable hypothesis, a measure, and a pre-registered failure criterion — Hypothesis-Driven Development. To date: 255+ hypotheses and 240+ experiments across 1,600+ expeditions, tracked in a graph-native system, with negative results published alongside the wins. Read the essay →

Contact: [email protected] · ORCID: 0000-0002-3912-0080