We publish the method, the measurements, and the things that did not work.
The programme has one position underneath it: Nothing exists in isolation, and a thing is the pattern its parts hold each other in. Each paper below is that position pointed at a different target — at machine self-models, at signals, at learning.
Everything here is open access with a permanent DOI, under CC-BY-4.0.
A machine's self-model must be powerless: it observes upward and can change nothing. Powerlessness is what makes the report trustworthy, and — the second finding — what makes a whole class of failure visible at all, because from inside a system a dead safeguard and a permissive one are the same observation.
The five commitments the work runs under, what each one rules out, and the one we learned by breaking it.
A signal is a curve of probability distributions on a statistical manifold. Comparing signals is comparing curves. Validated on speech across different voices and on handwriting from pen motion alone — including the failure case we predicted before testing.
What that distance does as a learner, with no trained parameters at all: 5–10× less data than a matched network, forgetting measured at exactly zero and provably impossible by construction, and next-step prediction 3–16× better than a trained network.
Papers are the summary. The working record is the honest version — 27 findings, of which 6 are negative and 6 are retracted, with the original wording of every retraction left visible and the reason it was wrong stated next to it.
It carries the mistakes that cost us the most: a distance whose apparent win was a padding artifact; a comparison run in a space where our own measurement was a constant function and could not have produced any other answer; a metric that compared a two-way choice against a ten-way one and made a growing label space look like memory decaying.
The repository holds seven deliberately cheating methods that our controls have to catch before any measure is trusted. On its first run, one of them got through, and we had to write another control.
Nothing here has been reproduced independently, outside the project that produced it. Every paper says so. Every page of the record says so. That is the ceiling on all of it, and it does not move until someone outside repeats the measurements.