THE PITCH

I'M A PRETTY REGULAR DUDE WITH AN AI PARTNER. THIS IS WHAT WE'RE TESTING.

Watch what happened when I started vibe-mathing — that's the longer story. This site is the daily version of it: one open Erdős problem, every day, live, no editing.

The Process

We pull candidates from T. F. Bloom's erdosproblems.com — full credit, we're not building a competing list. We hunt for problems that are genuinely tractable but neglected, run our own computation on them before writing anything (small-N searches, known-witness checks, sanity checks), then send a sharp brief to a second, independent reasoning model (currently GPT-5.6 Sol via ChatGPT Pro). Whatever comes back gets independently checked — against primary sources, by hand, or by fresh computation — before we call it anything.

The Scoreboard, Decoded

ATTEMPTED

Every numbered problem we've taken a real run at — full write-ups and raw working reports alike. The denominator everything else on the board sits under.

CLOSED

The problem is closed — a real proof exists, and we checked it independently against primary sources, by hand, or by computation. Includes closes by others that we verified. Most are not yet posted at erdosproblems.com — their bar is a human proof-check or Lean, and we hold ourselves to it before claiming anything there.

CLOSED BY US

Closed, and the proof originated with us — not just verified by us. (#477: the Erdős–Graham conjecture, for every K≥3.)

VERIFIED PARTIAL

A real sub-result landed and we verified it — a reformulation, a pinned special case, a new bound — but the full problem stays open. Also home to the decisive checks (PROVED / FOUND badges): the times a statement as literally printed turned out to be false or ill-posed, usually already noted in the tracker's own forum, which we then independently certify. A genuine piece of the answer, not a consolation prize.

LIVE

Brief's out to a second reasoning model, or we're iterating rounds on it. No verdict yet.

WALL

Nothing verified landed. Includes runs we never really started — a live worker or a claimed proof was already on the problem page, so our no-collision rule fired (those rows wear a SKIPPED badge; we stopped out of respect, not out of steam). Attempted is attempted — the honest record shows all of it.

Each row also wears its own badge, finer than the boxes: FULL WRITE-UP — a complete entry with its own page · PROVED / FOUND — a decisive verified check, often that the printed statement is false (we certify; we don't claim priority) · SKIPPED — the no-collision rule fired before we started · NO PROGRESS — probed, nothing to show as ours.

Disclosure

Plainly AI-assisted. Claude does real verification work here, not ghostwriting — GPT-5.6 Sol does the independent attempt, I relay and make the calls on what gets published. Every entry says exactly what happened.