x = f(x): ten pillars, their false premises, and the loop
A full whitepaper, produced under an unusual discipline: ten subagents drafted one derivation pillar each — gauge consistency, Noether, particle identity, vacuum structure, entropy and Bell, replicator-as-Bayes, Landauer, the good-regulator theorem, the LLM objective, kinetic proofreading — and nothing entered the paper until it had been re-derived by hand in the main session. A hostile proofread pass over the finished document caught four real errors before publication; the two seams where independent pillars touch the same physical number met flush to the last digit.
What’s inside: the axioms stated as axioms (A1–A10), the pattern-of-wrong table (epicycles→Fourier, caloric→Carnot, ether→Maxwell), all ten pillars with the math-vs-premise split applied to each, five figures, the spine chain from Shannon to the training objective, the Yoneda/Newman/Lawvere boundary section on what mathematics can and cannot deliver, and an exposure catalog — eight named ways the account dies, none of which has fired.
The split, stated plainly: every equation is theorem-grade textbook material, verified. The closure reading — that all ten pillars are one statement family, x = f(x) under one map — is axiom-grade: permitted by the math, not forced by it.
Companion post on the method: Ten agents and the loop.