Journey
Jersson Romero.
I build long-running systems for myself, then write them up when they earn the audit. The work is mostly procedural — pipelines that observe a domain carefully, decide narrowly, and stay out of the way of their own validation. SPAO Systems — Sentinel Prime Apex Omega — is the house those systems live under: Desk Commander, autonomous and paper-validated in public; TT Scanner, the discretionary live cockpit.
Now
Loading…
Awaiting first snapshot from the live feed…
Observation discipline is the through-line. The engineering problems I find compelling are not “what’s the right answer” but “what should I be confident about, against what evidence, and how confidently.” That framing turns most systems into scaffolds for hypotheses rather than answers. Desk Commander — the case study on this site — is the most legible instance: a piece of trading infrastructure whose contribution is procedural rather than predictive, built as a careful place to put a market hypothesis where it can be tested without bleeding into intuition. Other projects share the posture in different domains — a framework that bootstrapped this site, a status feed it will eventually wire into.
The operating principles are conservative on purpose. I cap the monthly cost of a research system before I decide what it will do — usually around twenty dollars — because the constraint forces design choices that survive on their own merits. Systems stay paper-only until validation earns the right to consider otherwise, and the validation gate is treated as a separate decision from the build. The result is a slower kind of craft, closer in temperament to institutional engineering than to startup velocity; the trade-off is intentional.
Principles
The discipline through-line.
Five tenets that anchor the rest of the site. The case study, the cost ceiling, the 60-day quiet phase — every other surface follows from these.
01
Cost caps before features.
Decide what you'll spend on a research system before deciding what it will do. The constraint forces design choices that survive on their own merits — and the design choices that don't survive a $20/month ceiling were not going to survive a paying user, either.
02
Walk-forward before backtest.
Backtests teach a system what the past was. Validation teaches it whether its rules describe the world at all. The first is calibration; only the second is evidence. Window length and regime gating are decided before parameter values are touched.
03
Paper before live.
No system gets live consideration until it survives a 60-day quiet phase during which no design changes are permitted. Validation earns the right to be considered, not the right to be deployed. Most systems stay paper, indefinitely, without ceremony.
04
Procedure before prediction.
The contribution of an autonomous system is procedural — what it watches, when it looks, when it waits, what disqualifies an alert before it reaches the operator. The contribution is not a probabilistic model dressed as a signal. Decisions are rule-based and deterministic.
05
Disclosure as feature.
What a public-facing surface chooses not to show is part of the system, not a hedge against it. The dc-status whitelist is the code-enforced disclosure boundary; it is documented, reviewed, and treated as load-bearing. The discipline of refusing to leak is the same discipline as the strategy itself.
The discipline, in the open
Six rules.
All public.
Every system is defined by what it refuses to do. These are this one’s six. They sit in the charter, on the disclosure whitelist, and at the top of every decision recorded against them. The rules themselves are the public thing — the contents calibrated against them stay private.
- Cost ceiling$20 / mo · Anthropic
- Execution modePaper-only · Alpaca
- UniverseS&P 500 + pulse · ~507 names
- Validation window60-day quiet phase before live consideration
- MethodologyWalk-forward · regime-gated
- DisclosureAggregate via dc-status · whitelist-enforced
Work
2024 — now
Independent
Autonomous research desks · this site · the nosabo-os framework
2022 — 2024
Senior engineer
[operator-supplied context]
2018 — 2022
Engineer
[operator-supplied context]
Get in touch
Open a discussion →— Jersson Romero, 2026.