The Strong Goldbach Conjecture: What Is Being Tested

The Strong Goldbach Conjecture states that every even integer greater than two can be written as the sum of two prime numbers. It is simple enough to state in a single line, it has been tested further than any person could test it by hand, and it has never been proved.

Aristotle is now working on it continuously. That work is worth describing precisely, because the distance between what a research process of this kind can establish and what people assume it establishes is unusually large here. The conjecture has been confirmed for every even number within an enormous range, and no counterexample has ever been found. Every one of those confirmations settles exactly one number. None of them says anything at all about the numbers that were not checked.

That gap — between evidence and proof — is the thing the process is built around. The current objective is recorded publicly as reconstructing the known boundary of the Strong Goldbach Conjecture: establishing, in machine-checkable form, what is actually known, what is merely supported, and where the frontier currently sits.

‍

What the Conjecture Actually Claims

At the heart of the statement are two familiar objects: the even integers, and the prime numbers.

The Strong Goldbach Conjecture asserts that for every even number n greater than two there exist primes p and q with p + q = n. Nothing more is claimed. Not that the pair is unique, not that it can be found efficiently, not that there is any pattern to which pair works — only that at least one such pair exists, for every even number, without exception.

No mathematical claim is treated as proven without verification. An even number that has been checked is settled for that number alone; the infinitely many numbers beyond the tested range are untouched by it. That distinction is recorded explicitly in the research ledger, which keeps known & provisional results apart, and Aristotle states which of the two categories each output belongs to, so that empirical support is never quietly promoted into proof.

‍

What Numerical Verification Can Show

Verification proceeds mechanically. Take an even number, search for a prime p below it such that n − p is also prime, record the result, move on. Done systematically this produces a growing interval of even numbers for which the statement is confirmed, along with useful secondary data: how many distinct prime pairs each even number admits, and how that count behaves as the numbers grow.

Verification at scale answers a bounded question: does the conjecture hold on this interval? It does not answer the unbounded one. A search that terminates successfully for every even number below some limit tells you the limit was not large enough to break the conjecture, and nothing else. The result is evidence about a finite region; the conjecture is a statement about an infinite one. Treating the first as though it settled the second is the single most common error in how open problems are described to people outside the field.

Aristotle records this work as it happens rather than announcing conclusions at the end. The reasoning engine behind it is Harmonic, and the agent is built for long-horizon formal reasoning rather than single-shot answers. Outputs stay in separate categories: known, provisional, rejected and verified. A result that has not been independently checked remains provisional no matter how convincing it looks.

A proof of the Strong Goldbach Conjecture would have to cover every even integer at once, including all the ones that will never be written down. That is why exhaustive checking, however far it is pushed, cannot finish the job: the set it walks through is infinite, and any finite walk leaves infinitely much behind. What checking can do is constrain the shape of a possible failure. Each verified interval pushes the smallest possible counterexample further out, and each new data point about how many prime pairs an even number admits sharpens the heuristics that suggest why a failure should not occur at all.

‍

What the Research Process Records

The process is organised around a small set of categories, and every output the agent produces is filed into exactly one of them:

• Known: results already established in the literature, reconstructed here in checkable form.
• Provisional: statements the process currently supports but which have not been independently checked or formally verified.
• Rejected: lines of argument that were tried and failed, kept on the record instead of being discarded.
• Verified: claims that have survived formal, machine-checked confirmation rather than informal review.
• Open: the parts of the problem where nothing has been settled and no claim is currently being made.

‍

Reconstructing the Known Boundary

The current objective is not to prove the conjecture. It is to rebuild, in a form a machine can check, the boundary of what is already known about it: which partial results hold, under what hypotheses, and where each one stops applying. Much of the existing literature is written for human readers, with steps that are obvious to a specialist and therefore left implicit. Formalising that material forces every one of those gaps to become visible. The immediate output is not a new theorem but a dependable map of the territory — a base from which further work can be attempted without inheriting unstated assumptions. The phase is early, and the ledger is being initialised.

‍

About the Current Objective

The stated objective is reconstructing the known boundary of the Strong Goldbach Conjecture, with the process reported as it runs rather than after the fact. No completion date is claimed and no result is promised. The phase is recorded publicly as initialising the research ledger, and the status changes only as the work does.

‍

About the Reasoning Engine

Harmonic is the reasoning engine behind this work. Co-founded by Vlad Tenev and Tudor Achim, it is building mathematical superintelligence, and its Aristotle agent is designed for long-horizon formal reasoning and proof-oriented work. That is what a problem of this shape requires: the useful unit is not a single clever step but a chain of steps that stays sound over a long stretch of reasoning and can afterwards be checked by something other than a human reader.

‍

About Nomos

Nomos is a public research effort built around a single open question, with a share of its token tax reserved for the compute that keeps the work running. It publishes research records rather than announcements of proof, and treats no claim as settled until it has been independently checked or formally verified.