D.E.C.
Request consultation

DavidEthanCarlisle

High-performance Rust engineering Local-first forensic systems
Request a consultation Accepting engagements  /  Q3 2026

Theory of provenance

“What’s in a name? That which we call a rose, by any other name would smell as sweet.”
Accepted William Shakespeare
Actual provenance 18th-century amalgam (Edmond Malone)
Source nodes Q1 1597 memorial reconstruction × Q2 1599 unedited manuscript

The most famous line ever written about naming is a structural fabrication. No printing in Shakespeare’s lifetime carried this text. It is an editorial collision — a patchwork merging an unauthorized memorial reconstruction with a manuscript riddled with unedited false starts.

In literature, a stitched-together reality is celebrated as canonical truth. In systems architecture, an unauthorized memorial reconstruction is a catastrophic failure of provenance.

When the origin of a data structure is ambiguous, the system is compromised at the root. Architecture demands absolute traceability. From hypergraph memory management to uncompromising duplicate detection, the objective stays singular: engineering immutable truth into high-performance systems.

Doctrine

Local first

Every system runs on the operator's own hardware. No account, no cloud service, no telemetry. Material that cannot responsibly leave a building does not have to.

Provenance preserved

Content is hashed on ingest and re-verifiable afterward. Where a record came from is recoverable at every stage, not inferred from a filename.

Nothing claimed unearned

Capabilities are stated at the level they were demonstrated. Where a method is lexical, it is not described as semantic. Where a result is bounded, it is not presented as proof.

Systems in operation

The forensic pipeline

Deployed / Rust · Tauri 2

Objective

Turn an undifferentiated mass of documents into an evidence base that holds up.

Scope

Three stages on one corpus. MinusOne scans terabytes of storage, hashes every file with streaming SHA-256, and removes duplicates. C4 compiles what survives — ingest, chunk, entity extraction, TF-IDF embedding, UMAP projection, HDBSCAN clustering, hypergraph construction — into a navigable three-dimensional atlas, reading MinusOne's index directly so verified work is never repeated. Dates, amounts, names, case numbers, and organisations surface as typed entities. Attached material re-verifies as intact, modified, or missing, and case exports ship with a manifest and a checksum file.

Built for

Discovery review, investigative research, and any archive whose contents cannot be uploaded to someone else's infrastructure.

RIGOR

Released / Assurance

Objective

Make an assurance claim impossible to state unless it was actually produced.

Scope

A counterexample-driven loop over a write-ahead log. A bounded reference model explores the state space and emits a counterexample against a deliberately broken recovery policy; an adapter validates that trace and generates a Rust regression test from it; the defect is reproduced under an explicit feature flag; the corrected implementation is verified against the same test. Only then does the sealer emit a proof-carrying package. It fails closed when evidence is absent or unsuccessful, and the sealed artifact records its own limits — unsigned, unanchored, bounded — in the evidence itself.

Includes

TLA+ specification, deterministic bounded exploration, SHA-256 source and evidence manifests, optional Apalache and Kani targets that are never claimed unless separately recorded.

Supporting systems

Shipped / Local first

LTE — Lexical Traceability Engine

Sentence-level provenance for text archives and source trees. Deterministic SHA-256 anchors, token frequency and co-occurrence, local Merkle integrity seals, and a read-only localhost API.

QPKM

Knowledge management over ordinary Markdown vaults rather than a proprietary database. Tantivy full-text search, global knowledge graph, atomic saves, timestamped local backups, recoverable deletion.

FluxRadio

Local audio curation and static station publishing. SQLite-backed classification, standards-correct HLS export, and a self-contained player that runs on any static host.

Otherplace

Local-first AI desktop. Ollama chat with multi-model broadcast, provider adapters, a researcher/drafter/critic pipeline, and a spend dashboard.

Terms of engagement

Three ways to bring this work onto your problem.

Mode I

Architecture review

A fixed-scope audit of an existing system: data structures, provenance guarantees, and the performance ceiling you are actually up against. Delivered as a written assessment with prioritised remediation.

Fixed fee
Two weeks
Written deliverable

Mode II

Build engagement

Direct construction in Rust — scanning engines, hypergraph structures, storage intelligence, deployment infrastructure. Scoped against a defined outcome rather than an hourly meter.

Project fee
Six weeks minimum
Shipping deliverable

Mode III

Advisory retainer

Ongoing architectural counsel for teams carrying a system they cannot afford to get wrong. Standing availability, structural review at decision points, no ticket queue.

Monthly retainer
Quarterly minimum
Limited slots
Request a consultation Book a call Response within two business days

Systems in operation

Not screenshots. The deployed systems themselves, running here.

Live · ghostofatlanta.com

Cityscape visualisation

Ghost of Atlanta

Rendered urban geometry driven by a narrative data layer — the visualisation and the story share one source of truth.

Live · humanbecomings.org

Interactive tour

Human Becomings

A navigable planetary sequence built to hold long-form argument without the interface competing with the reading.

Capture · Rust / Tauri

Desktop application

MinusOne

Duplicate detection across terabytes, with a streaming SHA-256 evidence chain that re-verifies as intact, modified, or missing.

Capture · Hypergraph

Compiler · Three.js

C4

Ingest through hypergraph in nine stages. Entities, clusters, and typed edges rendered as a navigable atlas.

Research

FluxFrame Publishing / Structural Linguistics Division

Toward Reduction Beyond Abstraction

Perception, evaluation, and the relocation of constraint. Argues that the recurring failure in machine cognition is not abstraction itself but antecedent abstraction — requiring a system to inherit its operative ontology before it has learned which distinctions matter. Proposes the Relocation Principle: constraint is relocatable, never eliminable.

July 2026  /  Read the paper →

New business inquiries