Tool-mediated workflows, bounded execution authority, deterministic evaluation, human/target control, and explicit failure states.
CAREER TECHNICAL PORTFOLIO · PUBLIC-SAFE
JONATHAN CHADBOURNE / FOUNDER & CHIEF ARCHITECT
I build systems that make consequential AI actions
harder to get wrong — and easier to verify.
My work at JCEE Labs sits at the intersection of applied AI, distributed systems, execution reliability, evidence, verification, and technical commercialization.
DALLAS, TXSUPPORT@JCEELABS.COM
TECHNICAL RANGEPUBLIC-SAFE SUMMARY
What I work across.
Idempotency, crash/retry recovery, concurrency, stale-state rejection, reconciliation, and durable effect histories.
Typed evidence, provenance and lineage, deterministic verifiers, adversarial testing, mutation campaigns, and conservative claim boundaries.
Customer discovery, solution architecture, demos, objective pilot acceptance criteria, SOW structure, and product boundary design.
Core source repositories, detailed schemas, hostile-test corpora, unpublished implementation methods, and partner-specific materials remain private for IP and trade-secret protection. This portfolio shows architecture, engineering decisions, controlled evidence, and claim boundaries without disclosing enabling proprietary detail.
SYSTEMS MAP
From proposal to independently checkable effect.
Proposal / requested action → authority & currentness → evidence closure → consequence boundary → durable effect execution → independent verification.
JCEE VOW focuses on execution assurance; QCS 2.0 on authority/currentness and transition legality; JEC + JCEE Assurance on portable evidence and independent verdicts; and Agentic AP Gate is the current product architecture applying those responsibilities to AI-proposed vendor-payment workflows.
CONTROLLED EVIDENCEBOUNDED CLAIMS
Measured work, with the ceiling stated.
tests gated by the JCEE VOW 1.1 private-release CI workflow
QCS concurrent PostgreSQL write attempts
unsafe commits in that controlled concurrency campaign
combined remote-authority QCS fault trials passed
JA-P0.3 adversarial mutations rejected or narrowed
false PROVED and false permission in the JA-P0.3 kernel campaign
These metrics describe controlled internal test and reproduction records. They are not claims of universal correctness, production certification, regulatory compliance, or coverage of every authority and failure model.
CASE STUDY 01
Reliable external effects under ambiguity.
I designed and tested private execution-assurance machinery around stable effect identity, durable journaling, recovery state, retry/refusal/observe-first behavior, and stale-authority rejection at ambiguous external-effect boundaries.
QCS 2.0 was independently reproduced against PostgreSQL transactional authority and a genuinely remote network authority. The controlled PostgreSQL campaign recorded 800 attempts with 0 unsafe commits; combined remote-authority fault trials passed 18/18 with 0 unsafe duplicate remote effects.
CASE STUDY 02
Evidence that can be independently checked.
JEC defines public-safe evidence responsibilities around identity, provenance, lineage, stream ordering, integrity commitments, and profile-bound validation. JCEE Assurance uses a separate deterministic verdict path to rederive bounded conclusions from retained evidence and trust inputs.
Controlled validation included 30/30 frozen interface cases, 14/14 native cases, fresh-process repetition, and 500/500 adversarial mutations rejected or narrowed with 0 false PROVED and 0 false permission in that campaign.
CASE STUDY 03
Productizing consequence control.
Agentic AP Gate is a current product architecture for the boundary between an AI-proposed vendor payment and an irreversible provider effect. The AI can propose; it is not given direct effect authority.
The architecture is designed around current-state checks, explicit consequence boundaries, durable effect handling, target-owned control, independent verification, and hostile/recovery acceptance scenarios. It is an authorized construction program, not a claim of completed live money movement or production non-bypassability.
ENGINEERING METHODFALSIFICATION-FIRST
Versioned, reproducible, failure-preserving.
Define the candidate, claim boundary, fixtures/corpus, acceptance criteria, and failure conditions before the run.
Exercise concurrency, mutation, stale state, crash/retry, replay, wrong identity, missing evidence, and bypass cases.
Retain failed experiments as evidence and distinguish semantic failures from infrastructure faults.
Change only the verified defect, version the successor, and preserve the prior candidate.
Use different substrates, separate verifiers, new implementations, and fresh-process repetition where possible.
Record exactly what the result establishes and what remains outside scope.
PUBLIC RECRUITING DEMONSTRATOR
Inspect the engineering style without opening the proprietary vault.
A small public recruiting demonstrator was authored from scratch to show a generic proposal → deterministic gate → simulated effect → receipt flow. It includes invented sample receipts, a deterministic test suite, and public-safe diagrams. It is explicitly not a subset, port, or reference implementation of protected JCEE systems.