86 cells, 8 states, one local rule. A loop of sheathed wire with a stream of instruction signals circulating inside it — its genome. Watch what the genome does: it gets interpreted (the signals run out an arm and build a body) and it gets copied (the same signals duplicate into the offspring). One strand, two readings — the move von Neumann saw was logically necessary for anything to reproduce, and the move DNA uses every time a cell divides.
Watch for it around generation 150: the arm reaches out, turns left four times, closes a new loop, and the umbilical cuts — then both loops start reproducing. The colony spirals outward and the inner loops freeze into a dead core, because a loop can't build into space another loop already occupies.
The signals circling the loop are a single instruction stream. When the stream reaches the open end of the construction arm, two things happen to the very same signals. They're interpreted — a grow signal extends the arm one sheathed segment, a turn signal makes it bend left — so the genome is being executed, read as commands that lay down a body. And they're copied — each signal that passes the T-junction is duplicated, one copy continuing around the parent, one copy sent down the arm to become the offspring's genome — so the genome is also being transcribed, read as pure data to replicate.
That's your Möbius, made literal and honest: "instruction" versus "data" is not a property of the signals — it's the same stream, and which one it is depends only on how it's being read at that moment. The two-ness lives in the reading, not the tape. (Where the Möbius image slips: there's no single half-twisted pass here — there are two distinct reads, interpret and copy, happening at the junction. One substrate, two readings — not one continuous twist.)
A genome that builds the body that copies the genome that builds the body. The loop you're watching is that sentence, running.
Langton got from von Neumann's 29 states, 130,622 cells down to 8 states, 86 cells by giving one thing up: universality. Von Neumann's machine could build any automaton from a description — reproduction was just the special case where the description was its own. Langton's loop builds only one thing: itself. So it copies endlessly, but it can't construct anything more complex than itself, and it can't meaningfully evolve.
That's the line that mattered three steps back. Copying is cheap — this 86-cell pattern does it. Open-ended growth of complexity is the hard threshold, and Langton's loop sits just below it: a perfect copier, frozen at one rung. Von Neumann's heavier machine sits above it; later variants like Sayama's evoloop cross over and actually evolve. The question worth asking of any self-propagating system — a cell, a worm, an agent — is which side of that line it's on.