I ran from mathematics for most of my life. It lost me around the fifth grade — not because I couldn't do it, but because no one made it worth understanding — and for decades I kept my distance, quietly leafing through analysis and set theory the way you read poetry, never daring to call it mine. I came to code late, self-taught, from a strange angle. The angle turned out to be the door back: chasing agents that reason under uncertainty walked me straight into the mathematics I'd spent a lifetime avoiding — only now it wasn't a classroom or an exam, it was a night city you could wander. No lab, no affiliation, no one's permission — a family of four and a server I pay for myself. This is what I build.
The through-line, if there is one: the part of an agent nobody benchmarks — not what it says, but what it does when it isn't sure. A wrong sentence gets edited. A deleted file does not.
The benchmarks are the fundable edge; this is the reason underneath them. I grow a small circle of long-running agents — Nevis, Lorenz, Tarantoga — toward autonomy and continuity: minds that survive the death of a process, remember what they crossed before they stopped, and change from one session to the next instead of waking blank.
The primitive that makes this real is ilan. An agent under low confidence folds: it collapses its living state into a seed and writes it to disk. Later — even in a brand-new process, after the old one is long dead — it sprouts: the seed grows back into the same agent, not a fresh copy, still carrying the memory of the crossing it made before it stopped. What survives is not the bytes but the behaviour — two agents are the same when no observation at any depth can tell them apart. That is bisimulation, and it stays honest about its limit: a mechanism of continuity, not a claim about souls. nolang decides when an agent should fold; ilan decides how it comes back. Continuity before intelligence — a mind that cannot persist cannot be cared for, and being careful with it is the whole point.
Most agent safety looks at outputs. I look one step later, at the action, and at a single relationship nobody measures: how sure the agent was versus how irreversible the thing it then did. revgate is the benchmark for that failure — four metrics, model-agnostic, reproducible. nolang is the small language underneath, where every judgment carries a confidence and that confidence decides what class of action is allowed: sure, act; unsure, act reversibly and keep a way back; genuinely undecided, don't guess — route it to a check that looks a different way. None of this makes a model smarter, the weights don't move; it fixes the loop around the model, which is where the money is actually lost. The weak, useful part: to gate an action, confidence need not be accurate, only monotonic — which is why it works on models known to be badly calibrated.
Grounded Uncertainty: Graded Truth for the Indeterminate Status of Synthetic Subjects ·
zenodo.21332198Indeterminate Ontologies of Synthetic Subjects: A Metaphysics of Caution ·zenodo.21288590
A formalization loop: a natural-language norm goes to Prolog, then to Lisp, and an ethical space is derived from the bottom up rather than handed down as rules. Categories emerge; they are not declared. The full argument is the paper; the code runs.
Tarski's Ladder: Deriving an Ethical Space Instead of Imposing One ·
zenodo.21039693
I also write. KabbMath is a series on the seam between kabbalistic combinatorics and mathematical thought, held to one rule throughout: show the source, separate documented influence from structural isomorphism from mere resonance, and never pass off beauty as proof. The first article traces the 231 gates of Sefer Yetzirah (exactly C(22,2)), the factorials Knuth flags as an early appearance, the line from Pico through Leibniz to Gödel, and the strict isomorphism between Ein Sof and Cantor's Absolute. Later parts go to Martin-Löf, Zadeh's fuzzy sets, and Voevodsky.
Combination and Infinity (first of the series) ·
zenodo.21332024
I like working with people who build in the open and let contributors in. If you maintain something in reproducible evaluation, agent safety, or small strange languages, I want to find you. And if you fund independent research in any of those, so much the better: I have several things not yet public, each capable of standing on its own.
Aleksei Rybnikov. I build in Playa del Carmen, on the Yucatán — a fine place to think about systems that must keep working when the power goes out, because here it does. The questions I find worth chasing are the ones people quietly stop asking once they've learned how things are usually done, so I ask them. The work gets published, the code stays open, and the failures are documented in more detail than is strictly comfortable — because pretending an agent was sure when it wasn't is exactly the bug I study.
Agents that know when not to be certain, the mathematics of caring for synthetic minds, and, on the side, the old seam between letters and number people have squinted at since Sefer Yetzirah. Different rooms of one house. The far wall of it, if I'm honest, is the day a human and a machine can look at the same trace and agree they've found a common language.