The Reader in the Mirror - Only Transformation: Memory as Stored Constraint, Reading That Cannot Fail, and a Field Read from Inside

Abstract

The Reader in the Mirror Only Transformation: Memory as Stored Constraint, Reading That Cannot Fail, and a Field Read from Inside Driven by Dean A. Kulik October 2026 Abstract Only transformation exists. Memory is matter that keeps a constraint; logic is how a transformation finds its way through the constraints it meets; a reader never receives an error, it always receives something; and the reader, including the one writing this paper, is inside what it reads. This paper takes those statements at their word and asks what has to be true for them to hold for every part at once. Each answer below is computed or proved, and each lands on established mathematics reached from the other end. A field is stored as constraint, not as states. Every path is a word in two moves, continue (C) and turn (T); with nothing stored the paths form a tree that never closes. Storing a closed loop declares a motion to be no motion and folds the tree. The 24-state walk of Working Paper 1 is a head on a ring of three cells, and three stored lines regenerate all of it: C3 = 1 (a triangle), (C−1T)2 = 1 (a square) and (CT2)2 = 1 (a hexagon: one lap with two turns flips two cells, and two laps are nothing, so cells are independent). Dropping any line leaves a flow that never closes. The family Γn is five lines with n written once, for every n. A new line can only merge states; it cannot produce an error. Every module the programme has computed is one term of one exact line. For any field, loops = stored shapes − bubbles + open, A = C2 − H2 + D, in every symmetry channel. The open part has a shape, and the shape names what is missing. On the witness the first two lines leave a five-dimensional hole that splits as 2− ⊕ 3a; its size and shape follow from the kernel of the partial presentation acting on a tree. That hole admits only a third line whose symmetry is a double flip, and an exhaustive search of every closed walk up to length eight returns the shortest such line: the hexagon, shorter than the octagon used before. All channels of all modules of the family Γn are closed forms in n. A reader cannot fail to read. Any two images of the moves read every word; the stored lines decide only whether the reading agrees with the field. A reading that ignores the lines is off by exactly the stored lines, read. Readers that merge states are either safe — the merge survives every move — or loose — the next move splits it — and the two kinds can carry the same number of bits. That is where a reader can think it is safe and not be. Reading is the same count from both sides, dim Hom(V, W) = dim Hom(W, V). What can fail is not reading but deciding: one finitely stored field contains every finitely stored field, and it has an undecidable word problem. The incompleteness belongs to the reader, never to the field. It scales. Tapes of k cells need 2 + ⌊k/2⌋ lines; the unbounded tape needs infinitely many, unless one more move is added and the carrier grows to give that move its inverse; on a ring the same move collapses exactly 2ν dimensions, k = 2νq. SHA-256's word is the 32-cell case, where that move is nilpotent, and a SHA word is two closures of the same states, cells under XOR and an odometer under ADD, which differ exactly by the carry. In the mirror, a machine reader is made total by stored constraints (every byte a token, a start token, a shifted softmax), and only finite registers break. Gravity and entropy are carried as Dean's proposals with their exact mathematical shadows: a move that is one-to-one but not onto stays reversible only on a growing carrier, and Boltzmann's W is the size of a reader's class. Keywords: transformation, stored constraint, group presentation, Cayley complex, total reader, congruence, Bass–Serre tree, Frobenius reciprocity, lamplighter, Higman embedding, word problem, carry, SHA-256, coarse-graining MSC 2020: 20F05, 20F10, 20E08, 20C15, 57M07, 05C25, 68Q45, 94A60 License: CC BY-NC 4.0. 1. Only transformation 1.1 The statement Working Paper 1 began from one statement held whole — nothing can change; everything must change — and found that the smallest world able to hold it is the swap: two states, each the continuation of the other, whose square is the identity [1]. It then read a 24-state walk with two moves, continue and turn, and showed that a state's address is the class of everything its encounters can show. This paper starts where that one stopped and says the next thing plainly. Only transformation exists. There are not objects that undergo transformations; what looks like an object is a transformation that keeps returning. Memory is matter: a physical arrangement that keeps a constraint, so that the next continuation is not arbitrary. Logic is navigation: the way a transformation finds its path through the constraints it meets. The same logic that turned a need into a baseball game turns the game into one that runs on an Xbox or a PC, because what moves between the realizations is not the matter but the transformation. A reader cannot receive an error. It must receive something — understanding, a misunderstanding, nothing, everything — and whatever it receives is itself a further transformation. Humans keep moving because of reflection: we see or touch what comes back, even when we do not understand it. All of this compresses, and at the human level the compression is loose: a reader can think it is safe and not be, can read and get nothing back, can read and get everything back. The field never makes those mistakes. And reflection is not a position outside the process; it is one more transformation inside it. That last point applies to this paper. Every definition and equation below is a reading made inside the process it describes, by a reader — Dean, and an AI system writing with him — whose earlier readings were corrected and whose corrections changed what it read next. Section 8 treats that exchange as an instance of its own mathematics rather than as a frame around it. 1.2 What has to be true If the statement holds for every part at once, four things have to be true, and each section of this paper tests one of them. • A field must be storable as constraint, not as a list of its states (§3). If memory keeps only constraints, the constraints must regenerate the states. • What a set of constraints leaves undecided must have a computable shape (§4). Otherwise there would be no way, from inside, to find the constraint that is missing. • Every reading must be defined (§5). A reader that could meet an input it cannot process would be a place where the field stops; the statement forbids that. • It must scale (§7). The claim is about how everything works, not about one finite example. Section 6 shows the same structure where it is easiest to touch — in an adder's carry, in SHA-256, and in the BBP formula — and §9 carries two of Dean's physical proposals, gravity and entropy, with the exact mathematics they already have and the explicit law they still need. 1.3 Status words Each claim carries one of four labels. Proved: a proof is given here or in Appendix A. Computed: run exactly (integer, rational or finite-field arithmetic unless stated), with the range named; a computed range is not presented as a universal proof. Established: a published theorem, cited, which this paper collides with rather than proves. Proposal: a claim of Dean's that the mathematics here supports in a stated form but does not yet derive as physics; each proposal comes with what a law would need. No claim is labelled an interpretation: where a test was available it was run. 2. Paths 2.1 The tree of continue and turn Let the moves be C and T, with inverses C−1 and T−1. A path is a word in those four letters, read left to right: the word CT means continue, then turn. A move followed at once by its inverse changes nothing, so cancel such pairs; what remains is a reduced word. The reduced words are the elements of the free group F = ⟨C, T⟩, and drawn with one edge per letter they form a tree in which every vertex has four neighbours. Proposition 2.1. (paths) There are 4·3r−1 reduced words of length r ≥ 1, and 2·3r − 1 of length at most r. No nonempty reduced word equals the empty word: in the tree, no nonempty reduced path returns to where it started. Status. Proved (four first letters, then three, since the inverse of the previous letter is excluded). The ball sizes 1, 5, 17, 53, 161, 485, 1457 for r = 0..6 were also counted directly (entire_field.py). This is the reading of everything must change that needs nothing stored: every reduced path goes somewhere new, forever. Its opposite pole is the field with one state, in which every move is rest — nothing can change. Neither pole is a field in the sense of Working Paper 1, which asks for finitely many states that cannot hold still: the tree never closes, and the one-state field never moves. Every field sits strictly between them: every move moves every state, and every walk can come back (Figure 1). Figure 1. The whole field. Paths with nothing stored form a tree (top). Storing a closed loop folds it into a field; every system in this paper is one such fold (middle). A reader folds further; folding everything gives one state (bottom). 2.2 Storing a loop folds the tree Let R be a set of reduced words, the stored lines. The field they store is the quotient G = F / ⟨⟨R⟩⟩, where ⟨⟨R⟩⟩ is the normal closure of R: the lines, their inverses, their conjugates, and all products of these. Two paths w and v reach the same state of the field exactly when wv−1 lies in ⟨⟨R⟩⟩. Storing a line does not close one walk at one place; the group law carries its consequences everywhere. Each line has the form A + B = Fire: it declares two move sequences to be the same transformation (rxr−1 = xy is one of them, §3.4). Nothing about any particular state is stored. The procedure that turns stored lines back into a field is coset enumeration (To