Boundary permeability: what a leak between realities would look like

The picture most people carry of two realities touching is dramatic: a portal, an explosion, something crossing over. That is the movie version. If branches of a multiverse are computational and imperfectly isolated, the physics of the boundary says the failure mode is quieter and stranger. Nothing crosses over. The isolation itself degrades, and the degradation has structure.

What permeability means here

A branch, in this programme, is a rendered environment running on a shared substrate. Perfect isolation between branches is an idealisation, like a perfect insulator. Permeability is the probability that a record formed in one branch leaves a distinguishable trace in the state of another. That is the whole definition. It is a number between zero and one, attached to a boundary, not a door.

The question that matters is what that number does over time, and what a nonzero value would actually look like from inside a branch. The answer is not meteor showers of impossible events. It is correlation.

Why the leak would be patterned, not noisy

If the substrate carrying the branches has topological order, and Papers 4 and 13 make that assumption explicitly, then information cannot leak across a boundary as free thermal noise. The bulk constrains what can cross. Whatever gets through is mediated by the same operators, along the same channels, every time.

So the signature of permeability is not randomness. It is a low-entropy, patterned deviation: the same structured error appearing in two places that are not supposed to be able to correlate. Two branches record mismatched histories with correlated mistakes. An error distribution that has grammar. A coincidence you can name.

This is why the hypothesis is hard to test honestly. Ordinary correlated errors, from crosstalk or a shared environment, produce patterns too. A leak that looks exactly like standard decoherence is unfalsifiable, and I have no interest in defending an unfalsifiable claim. That is why the testable content was pushed into shapes rather than events, which is what Paper 15 does with the noise spectrum and Paper 14 does with blinded protocols.

The field, and its repair

Paper 3 models the boundary with a field whose expectation value sets the permeability. The early version had a genuine defect: written naively, the leakage probability was an exponential that could exceed one, which is to say it was never a probability at all. At a modest input it blew up to ten to the twenty-seven. The repair, posted later in this archive, is one line: pass the suppression through a logistic function so the output is bounded. It took an external audit to prompt the fix, which is the part that embarrasses me, and the part worth writing down.

After the repair, the physically interesting claim survives: permeability is instanton-suppressed. It is not zero, but it is exponentially small, which is why nobody has seen it and why any honest proposal must say what would make the suppression visible.

What this is and is not

This is a hypothesis about a failure mode of isolation. It is not a detection, and nothing in this programme has observed a boundary event. The claim has explicit kill criteria attached in the papers: if a blinded surface-code experiment with adequate power shows error correlations matching ordinary crosstalk and nothing else, the hypothesis loses its only testable edge.

What survives even then is the structural point, which is the reason the idea is worth a post. In any simulated multiverse, isolation is an engineering quantity, not a metaphysical one. Engineering quantities fail along channels, along modes, along the constraints of the substrate. If you take simulated branches seriously at all, you should expect leaks, if they exist, to be lawful. A lawful leak is something you can go looking for.

The next two posts step back from the boundary to the bookkeeping underneath it, starting with the conservation law that the whole framework leans on.