Building deterministic systems for high-stakes environments where "it works most of the time" is not an acceptable operating model.
I work at the intersection of reliability engineering, runtime assurance, AI safety, and distributed systems. The common thread is simple: make fragile systems observable, auditable, and constrained by explicit mechanics instead of vibes.
Modern AI and distributed systems are powerful, but they are often opaque and operationally brittle:
- Observability stacks depend on trained models that can misclassify or hallucinate.
- Recovery paths are too often "restart and hope."
- Audit trails are bolted on after the fact.
- Compliance evidence is difficult to reproduce under stress.
I build physics-first and deterministic systems that:
- Replace guesswork with mathematics: variance detection, second-order signals, and finite-state correctness.
- Make failure visible: forensic recording, transition validation, and replayable incident evidence.
- Preserve sovereignty: read-only observation, minimal privileges, and air-gap compatible deployment models.
| System | Role | Status |
|---|---|---|
| Coherence SRE | Variance-based anomaly detection for mission-critical infrastructure | v1.0.0, DOI: 10.5281/zenodo.18002927 |
| Air-Node | Finite-state agent validation and forensic incident recording | Prototype |
| Blackglass Dojo | Adversarial safety testing for AI systems and guardrail failure modes | Active |
| Sovereign Reliability Lab | Deterministic phase partitioning under stochastic network conditions | Research |
| Sovereign Embed | WASM embedding service for deterministic edge inference | Experimental |
The primary systems have been audited for promise-code alignment, edge case behavior, and regression coverage:
- Coherence SRE: deterministic rolling-window signals, with amplification edge cases verified.
- Air-Node: forensic trail completeness, with repeated violations preserved and tested.
- Blackglass Dojo: adversarial safety stack wired end-to-end, with GAMMA scenarios reproducible.
Each hardening pass produced targeted fixes, tests, and merged changes on main.
High-stakes systems need evidence, not optimism. The work in this portfolio is designed around operational properties that survive contact with real environments:
- Compliance-ready: NIST-aligned design language, deterministic behavior, and deployability in constrained environments.
- Auditable: systems produce forensic records and avoid black-box decision paths.
- Low-risk integration: read-only sidecars and minimal privileged access wherever possible.
- Practical resilience: signals are designed for operators who need to detect drift before it becomes an outage.
Coherence SRE: A Derivative-Based Anomaly Detection Framework for Mission Assurance
Archived at Zenodo, DOI: 10.5281/zenodo.18002927.
- Location: Aurora, Colorado
- Email: colemanwillis@blackglasscontinuum.com
- Interests: SRE, runtime assurance, AI safety, deterministic observability, distributed systems reliability