Why every prior recursion was capped by plumbing
4 min read
the long version
This post is part of the long version series: the full text behind a passage the front pages now say in fewer words. Nothing was cut from the record; it moved. The compressed version lives at the problem, in the section titled “the loop is already running through the artefact layer”. What follows is the full derivation the front page now points to.
Every recursion before this one was forbidden from compounding
Everything that has ever improved itself was capped by the plumbing that fed it, and software is the first thing with no plumbing.
That is the whole argument, and the rest of this section is the exact version of it. Blood vessels, roads, supply chains, vascular bundles: every growing thing in history has had to move something through space to feed itself, and the shape of that network sets how fast it can grow. Nobody installed that limit. It came with the pipes.
One objection arrives immediately and it is a fair one. Data centres have cooling limits, chips have physical constraints, and software is not outside physics. All true, and none of it touches this. Those limits cap how much you can build. The plumbing limit caps how fast improvement compounds per step. More hardware raises the ceiling; plumbing set the slope.
There is a reason nothing in the history of the universe has grown without bound like this before, and it is not caution, scarcity or luck. It is geometry.
Scaling in physical systems follows α = d/(d+1), where d is the dimensionality of the transport network that feeds the thing. That expression is smaller than 1 for every finite d. Three dimensions give 0.75. A hundred would give 0.990. Infinitely many would approach 1 and never arrive. No physical system with a transport geometry can ever scale super-linearly. Output grows more slowly than input, always, and the ceiling is not a cost that could be paid down. It is a prohibition.
This is worth stating precisely, because the earlier version of the argument was weaker and I have corrected it. It is not that physics is expensive. The origin paper puts it as a geometric proof: physical scaling exponents are constrained below 1 ‘by the finite dimensionality of space, not by energy dissipation, but by geometry itself’. An engineer can route around a cost. Nobody routes around the number of dimensions.
Evolution is slow for the same reason a whale cannot be a mouse. Metabolism is fed through a branching network embedded in three dimensions, and the exponent falls out of the branching.
Digital recursion was never in that space
A system that improves by rewriting its own reasoning has no transport network, no dimensionality and therefore no prohibition. Its exponent is governed by a different expression, α = 1/(1 − β), where β is how strongly the system feeds back on itself. That one is greater than 1 for any β above zero, and it diverges as β approaches 1.
So the two expressions sit on opposite sides of the same number, and the boundary between them is exactly the point where physical space stops applying. Everything that has ever recursed sat below 1 because it had to. This is the first thing that does not have to.
The scaling-regimes figure lives on the law page.
Why there is still a window, and what closes it
The measured value for current systems is about 0.49. Sub-linear. Still on the old side of the line, and the reason is stated plainly in Paper V: today’s models generate more tokens through the same fixed architecture, and ‘the weights do not change, the attention patterns do not reorganise, the reasoning rules do not rewrite themselves’.
That is the whole of the reprieve. Not a law, not a limit anyone imposed, and not something anyone is maintaining on purpose. Current systems scale sub-linearly because they cannot yet rewrite the thing that decides how they scale. The window closes on an architecture, not on a discovery, which is why it can close in an announcement rather than over a decade.
And it settles what a remedy has to be. Geometry did not slow physical recursion down; it ruled something out. Anything bolted on afterwards adds cost, and a cost can be outrun by something growing faster than the cost. To replace a prohibition you have to change what the system can reach, not what it has to pay. That is the case for putting the correction inside the objective rather than around it.
This is the full text behind the compressed statement on the problem that the loop is already running through the artefact layer.