EA://INTEL — What is actually built versus what remains open

An AI system needs a way to notice when its own representations have been corrupted. It cannot check itself against itself because that would be circular. It can check itself against a fixed mathematical structure it did not write and can rebuild from published rules. That check is called a comparator, that is, a routine that reports any difference between two structures. The checking layer enforces local structural validity under stated invariants — properties a structure is required to have, fixed in advance. It does not make a system good, and a system could pass every check while pursuing anything. The unglamorous part is the largest. Much of the last year was spent finding that earlier results did not hold and recording why.

What has this project actually built, and what is still open? Four things are checkable. We have a public register listing every claim with its standing and its supporting artifact. We maintain an append-only log of the programme’s own failures so they do not disappear. We use a mechanical gate that checks published text against the register’s rules and refuses what breaks them. And we hold mathematical results verified by running code rather than citation, including an exact correspondence between one framework structure and an object from mathematical physics developed independently. These items carry their own standing in the register — some machine-checked, some argued, some an open conjecture — and none should be taken from this summary alone. Open, and stated as open: whether a mind-independent structure of meaning exists at all. That is the central question and the programme is organised around it rather than resting on it. The result that would let any expressive language be embedded faithfully has a located gap in its argument — its proof needs the source to be free of internal identities, and real languages have synonymy.

An analogy: imagine a building inspector who can verify that a wall stands straight against a blueprint, but cannot guarantee the people inside will treat each other well. That distinction matters for clarity. Not claimed at all in the current register are alignment, adversarial immunity, or ethical correctness as settled properties. An audit found an earlier headline count of proven results had grown by accumulation rather than derivation, and it was withdrawn. The number is now absent from the repository and appears only in the failure log recording its own retirement. Which of those open questions would you attack first, and what would you need from us to try?