dynamicsystemsarchitecture.org
New: four papers, two new verification results, two new interactive tools, and Explore → Fresh Publications

Open lab notebook

"Architecture itself is a form of knowledge. When the structure is correct, reasoning becomes stable, generalizable, and measurable."

— Academic Researcher

Read more testimonials →

"Architecture itself is a form of knowledge. When the structure is correct, reasoning becomes stable, generalizable, and measurable."

— Academic Researcher

Read more testimonials →

The four real channels — S, D, I, C — kept structurally distinct

Every finding here links to the log that produced it. Nothing is asserted without a document showing the test, the failure, or the fix. Status tags are honest — pending means pending, not "soon." How this lab stays rigorous ↗

For reviewers and early-stage evaluators

Problem. Adaptive systems — education, recovery tracking, decision support — routinely collapse distinct signals (environment, internal state, interface, execution) into one channel. Once collapsed, reliable state estimation and honest history become impossible to recover.

Approach. The Non-Collapsible Four-Channel Architecture (NCFCA) keeps those channels structurally separate rather than merged. A formal proof chain (current: v3.7), a real serialized event kernel, and a 10,000-trial empirical contamination benchmark — 0% contamination isolated vs. 44.1% [95% CI 43.11%–45.06%] monolithic — back the core claim.

Current status. Formal proof chain through v3.7 · deterministic event kernel with real replay verification · working GED instructional system (Wayfinder) built on the same architecture · sports backtesting application (27 seasons, 7,276 real games) · a 12-week curriculum teaching the underlying reasoning method · provisional patent filed.

Next. Structured pilots, technical review, and early support to move from verified architecture into controlled deployments.

State-Separation & Determinism Audit — an independent verification engagement, honestly scoped →

Research areas

Grows as work is added. Same field names used on gregorylacefield.com — one taxonomy, not two.

Systems Architecture / State Estimation Formal Methods / Mathematical Verification Adversarial / Security Testing Software QA / Verification Engineering Applied Statistics / Quantitative Backtesting Educational Technology / Instructional Design AI Systems Evaluation / LLM Auditing built The Four-Channel Framework The canonical S/D/I/C reference every domain application points back to.

The Reasoning Chain

Observe → Research → Decide — a real, verified pipeline for tracking research state as callable functions, not narration.

built · verified The Reasoning Chain Verification Registry → Observation Kernels → Research State Kernel → Decision Kernels.

Docs

Reference material, not findings — vocabulary and current-state summaries other pages link into.

built Current Understanding growing Glossary built Master Kernel Library built Capability Library exploratory Fiction

Formal proof — Non-Collapsible State Separation

Full version chain, four snapshots, each reproduced completely and chained by metadata box. Start anywhere — each page links both directions. All versions →

Current · v3.7 · June 2026 Flagship proof — 10,234 words Theorem 3, Corollary 3.1, Propositions 1.3–1.4, Open Problems 1–10. Superseded · v3.1 Post-audit snapshot — 4,211 words Superseded · v2.0 draft D-separation proof introduced — 3,860 words Superseded · v1.0 · May 2026 Initial version — 2,978 words
Start here Research Roadmap The whole research program as one real chain, five minutes to understand it all.

Core architecture — NCFCA All →

ncfca_lib · built Module Status Honest state of every module — what's live, what's orphaned. built Full System Audit What's wired in, what's orphaned, and a real drift risk found by auditing the actual call graph. Security · built Laundering Vulnerability — Found & Fixed

Learning & Tools

Complete education platforms, built on the NCFCA architecture, ready to use now.

built · executed Pipeline Orchestrator 4 real bugs found and fixed, then run end to end for the first time — real output shown. built · executed Verification Registry A "verified" claim automatically flagged stale the moment a new invariant was discovered — run, not described. built · executed Governance Layer A real operator-precedence bug, fixed, with a concrete before/after case — plus what "governance" actually means across this lab. built · executed Core Kernel & Pipeline Infrastructure The state estimation kernel, the estimator interface, and the pipeline runner — run fresh, one real bug found. built Estimator Comparison Four-channel estimators vs. standard filters vs. the real published knowledge-tracing literature — checked against a real search. Threat model · built Evaluation Package

Verification & testing methodology

Component A · built Full Changelog 9 real fixes, chronological, including the Fix 5 self-correction. built · index Independent Verification Center Every bug, audit, and real attack tested against this lab's own work — one front door. built · reproducible State Contamination Benchmark 10,000 deterministic trials, 0% vs 55.9% contamination — full download, every assumption stated. built Backtest-Only Audit Full list of what's still only ever been backtested against its own construction. built Execution Kernel Audit Cycle 10-item hardening checklist, line-checked — 4 fixed, 1 partial, 5 confirmed still open. built Estimator Correlation Finding The 15-to-1 duplicate-estimator finding. built Test 5 — Tuned Final Report Reconstructed and verified. built Kernel Taxonomy 5 real families, 15 registered kernels — not a full ontology, earned structure only. built Infrastructure (source) document_metadata_box.py, kernel_registry.py, four_channel_formulas.py — the white-box proof itself.

Sports / quant backtesting All →

built Walk-Forward Methodology, Plain Terms Why "did it work on past data" is the wrong question — tied to the real, running composer. built NFL 2024 Backtest Results Matched to full floating-point precision. built · interactive NFL Strategy Backtest Dashboard 27 seasons, 7,276 real games, real closing spreads. Eight classic ATS strategies vs. the real breakeven line. built · interactive NFL Strategy Composer Build a real multi-variable strategy, backtest it in-sample, forward-walk it against seasons it never saw. Wilson-interval confidence, not a bare percentage. built · synthetic data, stated plainly MLB Backtest Analysis Four-channel framework on MLB run margin, walk-forward tested, numbers independently recomputed from raw results before publishing.

Education / Wayfinder

pending · Cover page 7 Years Teaching → Software Formalization AI as delivery and measurement system for a proven method.
pending GED Misconception Database
pending Calibration Forward-Walk Result The honest bug-found story.

Wayfinder — GED instructional system

A complete, real GED framework built to be read and used directly by an AI assistant, on request — point your own AI here for real, verified lesson generation.

built · executed Wayfinder Verified computation engine, prerequisite-mapped curriculum across all four GED subjects, real misconception database.

Dynamic Systems Engineering — the curriculum

A 12-week course teaching the reasoning discipline this whole site is built on — for human learners, and for an AI to study before explaining any of this to a visitor. First of a planned sequence of four courses.

active development Dynamic Systems Engineering Foundations 9 of 12 weeks live. Observation, causality, state estimation, modeling, cybernetics, knowledge engineering, AI collaboration, methodology, system design.

Domain exploration All →

Lighter pieces. This section grows over time.

built · real data Boston Housing Backtest Domain-independence test — same kernel logic, real 1970 Boston housing data, honest failure case included. built · exploratory Domain Applications The four-channel architecture mapped onto Industrial, Athletic, and Co-Pilot domains — real mappings, connected to the already-tested Test 4 result. built · real data Authority Handoff Prediction — Tesla Real NTSB crash data, real forward-walk result: 71.4% vs 42.3% baseline. Honest scope, full download.
pending Correlation Evidence Compilation Pay-gap-style applied analysis.

Team

Gregory Lacefield

Creator & Research Director
Dynamic Systems Architecture, NCFCA, Wayfinder

LinkedIn Personal Site

Robert Barisoff

Recovery & Community Partnership
Las Vegas Rescue Mission, Research Collaboration

LinkedIn

For coaching, consulting, or appointments: Contact via LinkedIn or gregorylacefield.com

Comments

We'll only use this to contact you if we reply.

Comments

We'll only use this to contact you if we reply.