Research

Research

My research investigates the boundaries between what a system can determine, what it can infer, and what it can independently verify.

Proposed — Not Yet Started Current direction

Deterministic Boundaries for Probabilistic Systems

Under what conditions can probabilistic systems provide measurable utility beyond deterministic approaches, and what architectural controls are necessary to preserve integrity, confidentiality, traceability, and governance?

The first proposed investigation compares four retrieval approaches over structured DICOM metadata — deterministic lookup, PostgreSQL lexical/full-text search, pgvector semantic retrieval, and a hybrid of these — to ask when semantic retrieval earns its keep over simpler, cheaper, and more auditable methods, and what it costs in structural fidelity and provenance when it does. This experiment has not yet been conducted; nothing on this page should be read as a result from it.

Prior Independent Research

Two earlier, independently scoped research programs — not originally designed under the direction above — asked a narrower version of a related question, each in its own domain. Their results and open questions inform the direction above; the direction above does not retroactively change what either program tested or found.

Phase I Closed

Evidence Packets — Autonomous DataOps Research

Representing otherwise-equivalent operational evidence as an Evidence Packet found no demonstrated material correctness advantage under RQ1's prespecified materiality rule (13/60 vs. 12/60; −0.017 against a ±0.10 threshold). A follow-on comparison of durably captured versus later-reconstructed evidence found a threshold-crossing but statistically thin advantage for durable capture (RQ2). Full numbers, qualifications, and frozen artifacts are on the program's own page — treat this page as a pointer, not the authoritative record.

Latest publication: When Better Evidence Doesn't Produce Better Answers

Full Research Record

Research Milestone Complete / Paused

DICOM Trust Boundary Research

Demonstrated, with published evidence, that a pre-persistence structural parser and policy layer can inspect and selectively transform real DICOM objects without decoding what a policy doesn't require, and that the resulting policy boundary holds across four escalating persistence contexts (bare process, container, Cloud Run, GCP Healthcare API). This is a completed engineering milestone on real, independently reviewed evidence — it is not a claim of production readiness, comprehensive DICOM coverage, or a certified de-identification product. Full evidence is on the program's own page.

Full Research Record

How These Programs Relate

Both prior programs, independently and before any broader program existed, ended up asking a version of the same underlying question: what can a data system actually establish about itself, and what should it refuse to retain or claim past that point? Evidence Packets asked this for pipeline-incident diagnosis — what a diagnostic process can determine from captured evidence, and where representation and preservation each do or don't help. DICOM Trust Boundary Research asked it for medical-imaging privacy — how much of an object must be interpreted before a system can decide what it's permitted to keep. Deterministic Boundaries for Probabilistic Systems generalizes that shared question beyond either original domain, to when a probabilistic method (like semantic retrieval) should be trusted over a deterministic one, and what governance a system needs around that choice. Neither prior program was designed with this generalization in mind — the continuity was recognized afterward, not planned in advance.

More

Professional Experience

Resume

Employment history and independent research, kept clearly distinct.

Get in Touch

Contact

Consulting, contract, and technical-leadership inquiries, including on this research direction.