Michael Darius Eastwood · Independent AI alignment researcher
Published
The advice sounds obvious: submit the papers. Peer review will settle the matter. The advice is not wrong; it is incomplete. It underweights the economics of non-affiliated submission and it overweights what peer review actually delivers. Both roads are being walked here; here is what each looks like from the position of an independent researcher.
Institutional affiliation is not just a courtesy line at the top of a paper. It supplies infrastructure editors read: the author's department, the co-signatories, the ethics committee, the standing reputation that determines how much desk-reject work reviewers are willing to do on a marginal paper. A submission without any of these gets fewer minutes of attention on average and a higher desk-reject rate for reasons that are not about the work. This is not a complaint; it is an observation about attention economics in overloaded editorial pipelines. The remedy is to ship material that survives without those signals, which is a considerable amount of extra work most independent authors decline to do and this programme has spent eighteen months doing.
arXiv is stricter for independent authors than for institutionally affiliated ones, but not closed. The moderation process asks for the submitter to be endorsed by an existing arXiv author in the relevant category or to demonstrate publication history that substitutes. Both approaches are in play here. The papers targeted for arXiv are Paper X on the beta greater than k co-scaling condition, the blinding-protocol methodology paper describing the sign-flip result, and the paper stating the register construction rules as a general methodology for tracking convergence. Each carries its preregistration in supplementary material; each carries the replication package where applicable. The timeline is public on the programme's status page and will move on its own schedule rather than under external pressure to hurry.
A conventional research programme would publish, then advertise. This one advertised, and the advertisement is what the reader sees. That inversion is deliberate. The evidence chain, timestamps, register, dashboard, retraction, truth gate, exists so that verification does not have to wait for peer review. A reader who wants formal review can wait for it; a reader who wants to check the material now can check it now, using standard shell tools and the sources cited in the register. Both are valid. What would be misleading is to withhold the checkable material on the grounds that a slower channel is preferable, when the checks are already possible.
Peer review is a filter this programme has not yet passed and does not pretend to have passed. It is a real filter; it catches errors public dashboards will not catch. The current work stream is closing that gap on the timeline the venues determine, not on a timeline the programme chooses. Until the review process resolves, the record's response to the objection is not to deny it but to state it precisely, on the evidence page, next to the other honest limitations. Verification does not require peer review to begin. Peer review remains what the record is walking toward.
From the book Infinite Architects: Intelligence, Recursion, and the Creation of Everything by Michael Darius Eastwood.