Evidence

Results that others have checked

Are you coming from a leaderboard or a list of solvers (Packomania, ESA, FlyWire Codex, IBM Ponder This, Erich Friedman)? I am an independent researcher. I am developing a deterministic, learning and evolvable computing substrate for building embedded, verifiable and auditable AI, whose every decision can be replayed and verified. I test building blocks of my project, with language models, on these public challenges: I obtained these results on accessible hardware (described in the research programme), and it is the leaderboard managers who accept and publish them, after their own checks. They show that an approach combining probabilistic models, structured search, memory and deterministic verification can produce controllable solutions; they do not prove general intelligence.

These challenges are a test bench for my project. The results below are not assessed by me: they are published by the leaderboard referees, who check every solution against their own rules. Before each submission, I check every solution myself. A leaderboard position is a dated state, not a permanent title: the date of each reading is given next to each figure.

The results at a glance

• Circle packings (Packomania)
• Circle packings (Erich Friedman)
• European Space Agency (SpOC 4)
• Fruit-fly connectome (FlyWire Codex)
• IBM Ponder This

Circle packings: the Packomania “csqv” table

The problem: place N circles of free radii in a unit square, without overlap, so that the sum of the radii is as large as possible. The Packomania table, kept by Eckard Specht for many years, lists the best known values for each N.

Circle packings: Erich Friedman’s Packing Center

Erich Friedman’s Packing Center publishes circle packings whose sum of radii is maximal, in a circle or in a rectangle. Each entry carries the name of the person who found it and the month.

European Space Agency: SpOC 4

As read on 14 October 2026, my public Optimize profile shows four rank-1 positions on post-competition leaderboards: three Luna Tomato Logistics problems and the “large” instance of the Keplerian Tomato Traveling Salesperson. The sum of these ranks is the best possible (1 + 1 + 1 + 1). This is a summary of those four leaderboards, distinct from SpOC 4’s official overall result fixed at the 30 June 2026 deadline.

Four problems, two ways of framing the task

Luna Tomato Logistics concerns matching and transfer decisions between sets of orbits and destinations. Keplerian Tomato Traveling Salesperson asks for an order in which to visit targets while accounting for orbital mechanics. The pages below describe each instance and result separately; any animated diagrams are explanatory and do not replace the ESA evaluator.

Fruit-fly connectome: four FlyWire Codex challenges

FlyWire Codex offers computing challenges on the fruit-fly connectome. They extend my research on the propagation of activity in neural networks and on mapping the connectome. FlyWire accepts solutions onto the leaderboard and states that it verifies submissions; the “Winner” title of each challenge is awarded at its official deadline.

IBM Ponder This, September 2026

Every month, IBM Research publishes a mathematical puzzle and the list of the people who solved it. In September 2026, the task was to find an arithmetic progression of 35 Loesch numbers whose last term is as small as possible; a “*” bonus rewarded a progression of at least 42 terms.