I find what kills a document before it costs you. Contract. Policy. Spec. Framework. Filing. If it's a document, I red-team it — and I show you exactly where it fails, with a paper trail back to the document's own text.
Engage: aionsystem.github.io — intake, pricing, published audits, terms.
I take no sides. I read what the document actually says.
Complex documents have failure modes their authors didn't see. Rigorous authors miss things too — nobody reads their own document as an adversary. I do.
- What I Do
- Current Flagship Application
- Proof of Method
- Engagements
- Sheldon K. Salmon
- DOI / Citable Record
- How to Engage
- Links
I build formal instruments — scoring engines, falsification conditions, and structured epistemic tagging — that locate undefined terms, missing enforcement mechanisms, exploitable loopholes, and structural failure nodes in documents before they go into production.
This isn't a proofread. It's an adversarial audit with a scored, falsifiable output and a citation back to the document's own language for every finding.
The underlying operation is a recursive extension of the premortem technique: rather than stopping after one pass of "imagine this failed, why?", it recurses past that point — isolating the single node whose failure breaks the whole system, decomposing it against documented failure history rather than hypothesized failure, and continuing until it reaches something irreducible.
Where this applies: contracts, governance frameworks, AI/data policy, technical specifications, compliance filings, and regulatory architectures — anything where a missed clause or an undefined term becomes a liability later.
Every claim I make carries an epistemic tag, so you always know exactly how certain a finding is:
| Tag | Meaning |
|---|---|
[D] |
Data — directly observed, measured, documented |
[R] |
Reasoned — logically derived from [D] evidence |
[S] |
Strategic — directional claim about future action |
[?] |
Unverified — open question, contested, or resting on unverified primary text |
Nothing exits without a tag. Nothing is presented as more certain than the evidence supports. If a system's underlying math or logic cannot be traced to a derivation protocol, it is flagged and halted.
Ocean Systems Bridge (OSB)
Scaling the AionSystem method from static document review to dynamic, runtime structural integrity for autonomous maritime systems and deep-sea resource extraction. OSB is a layered protocol stack that converts unamendable regulatory text (e.g., COLREG, ISA Mining Code) and unsettled science into declared, bounded, testable, and auditable claims. It does not resolve uncertainty; it makes uncertainty visible, owned, and monitorable.
Four independent structural reviews, run on live, high-stakes documents, on my own initiative — no client, no fee. All public and verifiable, except where indicated.
Regulation (EU) 2024/1689, General-Purpose AI Systemic Risk. 18 structural failure nodes across three independent passes — 7 S4, 11 S3. Unified thesis: the EU regulated systemic cognitive infrastructure using product safety mechanics designed for discrete products. Every downstream failure is a symptom of that upstream mismatch.
Survived six rounds of external hostile review by a 15-role rejection council. Zero fabrication. 12 quotes verified against primary sources. 6 external legislative claims confirmed.
Published: DOI 10.5281/zenodo.22135551 · SSRN 7362039
Foundational medical ethics. 21 structural failure nodes. "Do no harm" is undefined, unenforceable, and has no adjudication mechanism or technological update trigger. Proposes a Hardened 2026 Reconstruction with a six-category harm taxonomy, independent counter-expert review, and AI-surgical oversight.
Published: DOI 10.5281/zenodo.22267142
ISBA/29/LTC/8, the ISA's draft Regional Environmental Management Plan guidance, in active UN negotiation at the time of review. 52 structural failure nodes across four sections, including four critical gaps that leave the primary environmental protection obligation unenforceable from adoption. Every finding traced to the document's own provisions.
Published: DOI 10.5281/zenodo.21326719
Structural review of the governing Data Protection Addendum for a major healthcare AI scribe platform. Traced voice-pattern data through 45 CFR 164.514(b)(2)(i)(P) — which names voice prints as federally defined biometric identifiers — against the contract's delegation of biometric consent, notice, and deletion obligations to the provider organization, and business-associate liability under the HIPAA Security Rule since the 2013 Omnibus Rule. The finding: a structural gap between what the contract offloads and what the provider's own intake consent language was ever built to cover.
Every claim links to its primary source; unverified claims pushed by a source were explicitly declined rather than promoted. Independently reviewed and validated by a healthcare compliance executive. (Not published as a standalone DOI — client-adjacent, redacted, held as an internal proof point.)
Three fixed-scope engagements. Full detail on the site.
One document, up to 75 pages. The three highest-severity findings written up in full. Delivered in 5 business days.
One document up to 200 pages, or a set of related documents. Every finding written up. Compounding-gap analysis. Root-cause classification. Insert-ready fix language. Delivered in 15 business days.
Sovereign diagnostic pipelines, custom frameworks, scoring methodology and epistemic tagging protocol tailored to a specific document ecosystem. You own the framework outright.
Standard terms:
- Full prepayment before work begins; no Net-30, no crypto, no wire transfers
- If the whole document has no S3 or S4 finding, full refund — the report is yours to keep
- Strictly asynchronous: no discovery calls, no status meetings, no pitch decks
- Client-provided NDAs are themselves red-teamed before execution
- Documents processed in ephemeral environments with zero-retention AI endpoints; raw files shredded 30 days post-delivery
Full terms: Terms of Service · Rules · Privacy Policy
Role: Red-Team Framework Designer · Structural Integrity Analyst
Location: Evans Mills, New York
ORCID: 0009-0005-8057-5115
I didn't set out to specialize in one field. The interests stayed broad, and the through-line turned out to be a single reusable operation instead: find where a system fails, recurse past where most analysis stops, and ground every finding in real precedent rather than instinct. The frame I use for this is Advocatus Diaboli — the historical "Devil's Advocate" office in the Catholic Church's canonization process, whose entire job was to argue against a candidate's case regardless of who they were. Same structure, generalized: the target changes; the operation doesn't.
I reason in a persistent, spatial way rather than in prose or equations, working inside a self-built internal space (Loci World) built on the classical method of loci. On the standard psychometric measure of imagery vividness, I scored 78/80 (~95th percentile, hyperphantasia range) — a data point on how the thinking happens, not a claim about why the method works.
My standard applies to my own output first: if a framework's underlying logic cannot be traced to a clear derivation protocol, it is flagged and halted. Catching your own errors or epistemic drifts before they reach production is the job, not a liability to hide.
Affiliations: ARIA — Massively Scalable Neurotechnologies for Human Health Programme (Teaming Programme, Applicant, 2026–present) · Foresight Institute, Computation & Health Extension community (2026–present)
Full methodology, citations, and worked case studies: "Risk in the Risk in the Risk: Recursive Risk Discovery as a Polymath's Converted Function" (2026) — DOI: 10.5281/zenodo.22665348
Works registered and publicly archived through Zenodo. Full framework, brain-architecture, and world-building repos exist behind the scenes and are cited here for the record — the work above is the front door.
| Purpose | Cite This |
|---|---|
| EU AI Act audit (Arts. 51–56) — 18 failure nodes | 10.5281/zenodo.22135551 |
| Hippocratic Oath & Declaration of Geneva — 21 failure nodes | 10.5281/zenodo.22267142 |
| ISA Deep-Sea Mining Code (ISBA/29/LTC/8) — 52 failure nodes | 10.5281/zenodo.21326719 |
| Recursive risk methodology / cognitive profile, worked case studies | Risk in the Risk in the Risk |
| Organization as a whole | ORCID 0009-0005-8057-5115 |
If you need a document, contract, spec, filing, or governance framework red-teamed — the intake form is on the site: aionsystem.github.io/#contact. It builds a structured email; nothing is sent until you press send in your mail app.
If you are a skeptic — good. Test the method. Every case study above is DOI-linked and downloadable; the quotes are verifiable against primary sources. Find the gaps. Challenge the tagging with specifics.
AionSystem · Sheldon K. Salmon — Document & Framework Red-Teaming · Structural Integrity Analysis
I find what kills a document before it costs you. I take no sides. I read what the document says.





