The question everyone asks first · answered with numbers, and with what the numbers are worth
How likely is this, honestly?
A companion to the creation hypothesis. It scores that hypothesis, compares the score with the famous rival’s, and shows why the comparison does not mean what it looks like. Nothing measured on this site depends on any of it.
The honest answer to the question in the title is that nobody can give you a defensible number today, for this idea or for the famous one it gets compared with. That sounds like a dodge. It is the opposite: what follows shows why the famous number is not a measurement, and then builds the thing that can be measured instead.
The famous number is not a measurement
The simulation hypothesis is usually reported at around fifty per cent, sometimes twenty-five, sometimes seventy. Those figures come from serious people doing serious work, and they are not what a casual reader takes them for.
Probability updates by a simple rule: what you believed before, multiplied by how much better one idea predicts what you actually see. If two ideas predict exactly the same observations, that second factor is one, and multiplying by one gives you back what you believed before. You have measured nothing. You have restated your starting assumption in decimal.
Now recall what the simulation hypothesis says: a sufficiently good simulation is indistinguishable from the reality it simulates. Indistinguishable is the whole premise. So there is no observation that favours one side, the multiplying factor is one by construction, and the celebrated fifty per cent is the starting assumption coming back out unchanged, wearing the clothes of a result.
That is not a complaint about the simulation argument, which is a legitimate piece of philosophy and an ancestor this programme credits. It is a correction to how its number gets quoted. And it changes what my own missing number means: this hypothesis has never been scored because nobody has written down a starting assumption for it, which anybody could do in an afternoon. So here is one.
Why the ceiling is the assumption, not the world
The point is sharper than it first looks, and it can be shown with arithmetic anyone can repeat. Suppose a creator-type model produces N realities and there is one uncreated original. Counting observers, the chance of being in a created one is N/(N+1), which climbs towards certainty as N grows. But that is a figure conditional on the model being right. Average it against the possibility that the model is wrong, and the ceiling is not certainty; it is whatever prior weight the model was given in the first place.
| Created realities | Two families (creator prior ½) | Three families (⅓) | Six families (⅙) |
|---|---|---|---|
| 1 | 0.250 | 0.167 | 0.083 |
| 10 | 0.455 | 0.303 | 0.152 |
| 1,000 | 0.500 | 0.333 | 0.167 |
| a million | 0.500 | 0.333 | 0.167 |
The famous ceiling is a half because the calculation was set up with two families. Put natural cosmology in beside them and it is a third. Six families and it is a sixth. No astronomy changed between those columns. The number tracks the size of the list somebody chose to compare. That is why this page will not answer the title question with a percentage of its own: it would be the same trick with different furniture.
What can be built instead: a threshold, not a guess
There is a better question available, and it is the one the epidemiologists ask. Whether an outbreak grows or dies out does not depend on anybody’s opinion; it depends on whether each case produces more than one further case on average. Chains of universes work the same way. Write down the average number of viable daughter universes one universe eventually produces, and call it the reproduction number.
It factors into things that can be argued about separately: whether a universe grows intelligence at all; whether intelligence reaches the capability; whether making a real cosmos is physically possible; whether anyone who can, does; whether the daughter is viable; whether the lineage stays stable long enough to get there; and how many attempts succeed. That last-but-one term is where this programme’s measured work sits, because a lineage that cannot correct itself does not survive to make anything. It is placement, not evidence: those measurements do not yet raise or lower the cosmology’s odds, and this page does not let them pretend to.
What the maths then gives is not a guess but a hard edge. Below a reproduction number of one, the chain dies out with certainty, however many universes it starts with. Above one, survival becomes possible and rises quickly.
| Average viable daughters | Chance the lineage survives indefinitely |
|---|---|
| 1.00 or below | 0 |
| 1.05 | 9.4% |
| 1.10 | 17.6% |
| 1.20 | 31.4% |
| 1.50 | 58.3% |
| 2.00 | 79.7% |
| 3.00 | 94.0% |
The one place where evidence actually decides something
There is exactly one observation on the table that prefers one story to another, and it is the number quoted at the top of the hypothesis page: our universe began in a state of order so extreme that Penrose put the precision at one part in ten to the power of ten-to-the-hundred-and-twenty-three.
Compare three ways of explaining why we find ourselves somewhere so orderly, because leaving the third one out is how this kind of argument usually cheats.
Because that is simply how the universe began. No selection, no chooser: the initial condition is a brute fact, or one day a theory will show it was the only condition possible. This is the mainstream position and it is the one to beat. It is not scored below, because it has no free parameters to score: it wins by default if either of the other two fails, which is exactly why the other two must carry their own weight.
Because observers must exist to see it. This is the standard anthropic answer, and it has a known problem that is not mine: it explains far too little. Observers need only a modest patch of order, so this reasoning predicts we should find ourselves in a universe vastly less special than the one we are in. Penrose himself pressed exactly this point, and the same gap is what makes the so-called Boltzmann brain problem bite.
Because a universe that grows a maker needs more. Raising a mind capable of building things is not the same as housing an observer. It needs a thermodynamic gradient that lasts, because learning costs energy: every bit a system erases costs it at least a fixed minimum, which Landauer established in 1961. It needs deep time, because compounding needs generations. It needs stars and chemistry, because minds are built from something.
So the two stories predict different amounts of order, and the direction is fixed by logic rather than by taste: every universe that raises a maker also contains observers, and not the reverse. The maker requirement is strictly the larger one. Which means that on the first story extreme order is a puzzle, and on the second story extreme order is what you would expect to find.
And the prediction generalises past entropy, which is where it becomes properly testable. Selection for observers should favour constants that permit life. Selection for makers should favour constants that permit life and the long stable climb to building things, which is a stricter filter. So the distinctive prediction is that our constants should look enriched for that capacity beyond what mere habitability needs. Nobody has yet defined how to measure such enrichment; defining it is the work, and this page says so rather than claiming the prediction already succeeded.
Why the rival’s energy problem is not this one’s
In 2025 the astrophysicist Franco Vazza did the calculation the simulation argument had gone without, and published the bill in Frontiers in Physics. Simulating Earth alone, to begin with, needs energy of the order of a globular cluster’s entire rest mass. Advancing a simulation of the universe by one second needs operation counts no hardware obeying our physics can reach. His conclusion is that a universe like ours cannot be simulated by a universe like ours.
That result is fatal to the rival and irrelevant to this hypothesis, and the reason is a single line of arithmetic rather than special pleading.
The honest caveat belongs in the same breath. Instantiation is cheap only if it is possible, and possibility is exactly the term this page scored at one to ten per cent. So the trade is not a victory: the rival is expensive and possible in principle, this one is cheap if it is possible at all. Anyone who tells you their idea wins on every axis is selling something.
What would change these numbers, stated before the fact
Committing in advance is the part the famous rival has never done, and it costs nothing but nerve.
Upwards. Measured confirmation that correction keeps pace with capability in self-improving systems, which the programme is testing now. A laboratory demonstration that a vacuum bubble can be nucleated without the singularity that blocks it classically. A formal result showing maker-selection recovers the observed order where anthropic selection cannot.
Downwards. Any of the three killers above. A proof that making a universe costs what simulating one costs, which would collapse the cost asymmetry entirely. A demonstration that correction cannot keep pace, which ends the chain’s ability to persist and takes the whole reproductive picture with it.
The conclusion, which is not the one I would have chosen
This hypothesis is not more likely than the simulation hypothesis, and this page has not claimed a number for either. Neither is scored against the mainstream position, which needs no chooser at all and remains the one to beat.
What can be said is narrower and, I think, worth more. The famous rival has a starting assumption and no evidence that discriminates. This one now has a starting assumption broken into four attackable terms, one genuine piece of evidence pointing its way, a computable reason why the physical objection that damages the rival cannot reach it, and a list, published before the fact, of the findings that would move it in either direction.
Where next: the hypothesis itself, with its record lines and its seven kill conditions; the paper, where the same analysis is set out formally; and the dated sentences, if what you want to check is when any of this was first said.