A one-person research operation. The pipe is private. The outputs are public, dated, and auditable.
I am an independent researcher operating without institutional backing. The lab is the project. The Resolution Engine is the pipe inside the lab. The outputs sit across physics, finance, music theory, molecular structure, and AI security — each one dated, sourced, and posted in public.
The bet of this page is simple. The conversation about AI right now is mostly about the model. The actual lever is what you ask the model to do, with what context, against what private knowledge. The lab is a working demonstration of that thesis. Same frontier models everyone else has access to. Different pipe. Different inputs. Different outputs.
PROJECT FRAME · The Lab is the project. The Engine is the apparatus. The papers, tools, and registry are the evidence the apparatus works. All three are the same case.
The engine is a structural analysis suite. It runs problems through ten deterministic middleware stages (P05 → P50) against two private libraries of structural patterns. A frontier model appears only at the final stage — and only as a translator. The analysis is already done before the LLM is called. Built with a debt to Wanderland by Graeme Fawcett — the architecture that made the runtime legible.
A problem enters the pipeline as text. A deterministic compiler segments it into claims and matches them against pattern signatures via coordinate-space resolution on eighteen concept neighborhoods — not keywords, not embeddings. Subsequent stages compute tension between co-active patterns, search precedent diagnoses, run cross-domain transfer, tier the findings, and validate coherence. By the time the LLM is called at P50, it receives a finished structural diagnostic. Its job is to render that diagnostic into language. It does not analyze. It translates.
Twenty layers, Distinction through Dark Energy. Encodes the framework's base mathematical concepts — fold structure, closure cost, scaffold-to-3D projection — as nodes the compiler can bind to. Some layers are public Zenodo preprints. The full library is private and auditable.
Eighteen-plus structural patterns. Each pattern is a markdown node — YAML frontmatter, detection signatures, executable examples. Drop a new node into the directory and the compiler gains new capability. No code changes. The file is the feature. Some patterns are published; others are held as working IP.
AUDITABLE ON REQUEST · The libraries and the structural diagnostic any case produces are open to qualified counterparties under NDA. The pipe is private. The outputs it produces — section 04 — are not.
Every current conversation about AI capability is framed around the model. Bigger model, smarter agent, longer context window, better reasoning trace. The frame is upside-down.
The lever is what sits upstream of the model call. What you ask the model to do. With what context. Against what private knowledge. In what grammar. The model is the executor — important, but increasingly commodified across frontier providers. The thing that distinguishes one operator from another, at this point in the curve, is the input architecture.
The Resolution Lab is built around a stronger version of that claim. The engine doesn't just prompt better. It moves the analysis out of the LLM entirely — into deterministic Python that runs against the private libraries — and hands the LLM a finished structural diagnostic to render into language. The AI is the translator at the end of a ten-stage pipeline, not the analyst. That's why the outputs in section 04 hit across physics, markets, molecules, and tooling with the same machinery. The machinery isn't the LLM. The LLM is the last mile.
A selection. Each card is one result the pipe produced, a one-paragraph framing, and the links to verify it directly. The full Zenodo community has more; this page is the curated entry path. The point of the breadth is not the count. It's that the same machinery hits across physics, markets, molecules, and tooling. That's what makes it a grammar rather than a one-off.
The Koide formula relates the three charged-lepton masses by a clean ratio. Discovered empirically in 1981. Resisted derivation for forty-five years. Here it falls out of the two-axiom framework as the equilibrium condition for fermionic identity projected from a 3D fold onto a 2D scaffold. No fitted parameters.
A phase-coherence burst-energy dashboard for the S&P 500. Tracks topology-adjusted index level, burst Z-score, lead/lag regime, and five signature states. Live regime transition observed in real time — correlation flip anticipated before it completed. Resolution-dependent horizon boundary near 280–315 trading days.
A code-review tool that treats a repository as a graph and flags hot module-boundaries via Louvain community detection. Validated against rust-lang/rust with a 9.1σ hot-boundary finding — a structural seam the conventional review process had not surfaced. Deployed and operational.
The closure-cost grammar — C(p,q) = p+q−1, two axioms, no free parameters — applied to molecular structure. Recovers the Madelung rule as an identity-level hit. Identifies hubs at 9 of 9 test molecules. Decomposes ring strain into topological plus angular components. Cross-validated by a May 2026 PNAS paper on oligocarbamate binding (Escobedo / Alshammasi), whose binding-entropy mechanism maps directly onto the phase-loop topology.
Musical consonance derived from torus-knot closure cost. The chord-scale extension recovers Hugo Riemann's 1890s harmonic-dualism argument from first principles. The two-body case is the same machinery as the molecular grammar in the previous card — that's the point of calling it a grammar.
Schrödinger. Navier–Stokes. The heat equation. Black–Scholes. Each one derived from the framework's two axioms and the fold structure — not as a curiosity, but as a consistency check. If the grammar is right, the canonical equations have to fall out. They do. The papers are dated and on Zenodo.
A working lab is not a finished portfolio. Three threads in active research right now, each with a named next action. Dated. In motion. Falsifiable.
Closure-cost mathematics extended to atomic and molecular structure. v2 architecture specified. Four falsification tests defined. PNAS oligocarbamate paper provided strong cross-validation; binding entropy maps onto the phase-loop topology.
S&P 500 sectors framed as closure-cost resonance clusters. The cost-min to cost-max gap is hypothesized as a regime-diagnostic property. HCCI saturation is the agreed open diagnostic before new infrastructure is built.
The two-body case shipped in February (musical consonance). The chord-scale extension shipped in May. The molecular case is in flight. The claim under test: one closure-cost grammar applies wherever finite-alphabet closure structure exists.
I am running a working lab without institutional backing. The pipe is private, the outputs are public, the work is dated. Continuing it at this velocity is a resource question. Three ways the reader of this page can move the needle, in priority order.
The lab is running. Continuing it at this velocity requires capital. Open to grants, sponsored research seats, fellowships, patronage, or non-venture arrangements. The libraries are private and auditable; the outputs are public and dated. That's the diligence package.
Open a conversation: jasonrconnerty@gmail.com
If funding the lab independently isn't the fit, the alternative is a role that treats this work as core. Research roles, AI input-architecture roles, anywhere the section-03 thesis is the job — not roles that need a generalist who happens to publish on the side.
Filters · Atlanta-based or remote. Roles where the input-architecture thesis is load-bearing, not decorative.
If the first two aren't actionable, the lowest-bandwidth way to help is to read, share, and engage publicly. Discourse is what makes the first two more likely. Pick a card from section 04, follow it down, post the part that lands.
Where it lives · Zenodo · Medium · LinkedIn · Bluesky