For more than sixty years radio telescopes have listened to the sky and found nothing. No narrowband signal, no laser pulses, no waste heat from large structures. This silence is the most consistent finding in modern astronomy. It is usually treated as an open question.

Michael Garrett treated it as arithmetic in 2024. He holds the Sir Bernard Lovell Chair of Astrophysics in Manchester and directs the Jodrell Bank Centre for Astrophysics. Writing in Acta Astronautica, he took the Drake equation and reversed it. Instead of asking how many civilisations exist, he asked what lifetime we would have to grant them for our data to come out empty. Even under generous assumptions about how common life is, the arithmetic lands at one to two hundred years. At any given moment one or two transmitting civilisations would be active in the Milky Way.

This is not a prophecy. It is a back-calculation from an observation. If technical civilisations arise and we still hear nothing, then they do not stay loud for long. Something ends their loud phase, reliably enough that it happens to all of them.

Garrett names the candidate he considers most likely: their own artificial intelligence. I regard that attribution as open to debate. What occupies me in his paper is a different passage. It is the passage describing what would save us.

Why the Arithmetic Reaches Us

The core of his argument is a comparison of two speeds. Artificial intelligence advances along compute and data. Its progress costs electricity and time, but no tonnage. The ability to live permanently away from Earth advances along mass, energy and material limits. Every step has to be physically lifted, moved and operated in an environment that forgives nothing.

Garrett concludes that one curve overtakes the other. We are closer to a technical singularity than to genuine spacefaring capability. A self-sustaining civilisation across several worlds would take many decades, possibly centuries. He names the reasons plainly: energy limits, materials science, the harshness of the environment. He also points to unresolved questions in medicine, behaviour and logistics.

Here I agree with him without reservation. Anyone who has watched language models develop alongside landing vehicles over recent years has seen two clocks running at very different rates. One is read in months. The other in programme decades.

That is precisely why the next step of his argument deserves closer attention. Garrett does not stop at the diagnosis. He names a countermeasure.

The Threshold Nobody Quantifies

His countermeasure is distribution. A civilisation spread across several celestial bodies distributes its risk. If one site fails, the others survive. The single point on which everything hangs becomes several points that stand independently. It sounds compelling, and it is the underlying figure behind almost every justification of spaceflight that goes beyond curiosity.

The paper describes these sites with adjectives. They are to be self-sustaining, resilient, enduring. Nowhere does it say how one would recognise this. There is no quantity, no value, no condition a site would have to meet in order to count as insurance. The paper's most important claim is therefore untestable.

I propose the missing condition. I introduced the term in Civilizational Brief #153, under the title "The Unsupported Duration". A site insures nothing until its unsupported duration exceeds the resupply interval. Unsupported duration is how long an installation lasts when nothing arrives. The resupply interval is the gap between two arrivals. If the duration is longer, the site survives a failure of supply. If it is shorter, it ends with it.

A site that meets this threshold I call a fallback site. Anything below it is an outpost. The difference is not one of degree. An outpost extends the deadline. A fallback site carries it.

In Garrett's paper both go by the same name. At one point planets and outposts in space stand side by side, as if they were the same kind of answer. They are not.

What an Outpost Actually Is

Apply the threshold to what is being built or planned today and the result is sobering.

The International Space Station normally holds around six months of food and water. It is resupplied several times a year. When three cargo flights were lost in succession in 2015, the margin fell to four months. The revealing figure is the limit NASA named at the time. At forty-five days of supply, planning begins to bring the crew home. The answer to missed deliveries is therefore not self-sufficiency. It is evacuation.

Water recycling does not change this. On paper it nearly covers demand. In practice a deficit remains that has to be made up from below. The station's unsupported duration therefore sits in the same range as its resupply interval. A first-generation lunar base will not fare better. It brings its breathing air or extracts it from stores, its spare parts come from below, its medicine likewise. Even ambitious Mars designs assume supply windows that open every twenty-six months. Unsupported duration would have to exceed that span with margin, otherwise any delay is fatal.

From this follows something that rarely appears in the rhetoric of insurance. A failure on Earth does not reach such a settlement later than it reaches us. It reaches it first. Before anyone on the Moon notices a catastrophe, the launch campaign keeping them alive has already stopped. The colony then perishes not as a witness, but as the dependent link of a chain whose first link has snapped.

An outpost is therefore not a second chance. It is a second vulnerability with an umbilical to the first. Anyone booking it as insurance has read the direction of dependency backwards.

The Isolation That Isn't One

There is a second idea in Garrett's paper that reaches further than the first. He proposes using sites away from Earth as isolated environments. Advanced systems could be tested there without a failure taking immediate global effect. Different locations could host different stages of development.

This idea does not survive contact with physics, for a reason that has nothing to do with engineering.

Distance does not separate symmetrically. It severs control, not effect. A system on Mars can no longer be steered from Earth, because every query travels for between three and twenty-two minutes and the answer takes the same time again. Every two years the Sun stands in the way and the link drops entirely for roughly two weeks. During that period nobody on Earth decides anything about what runs there.

The output of that system reaches Earth regardless. A design, a model, an instruction carry no mass. They travel at the speed of light and arrive as soon as the channel reopens. Physical separation produces no informational separation here. It produces a situation in which something operates beyond our reach and takes effect within it.

What the paper describes as a gain in safety is a loss of control at unchanged reach. It is the worse half of both worlds.

The Second Site Is Being Built, Just Not for Us

While insurance is being written about, distribution is actually happening elsewhere. Only not the kind Garrett wants to see.

Axiom Space brought the first two orbital data centre nodes into service in January 2026. They run on the optical relay network of Kepler Communications. Starcloud has operated a high-performance GPU in orbit since late 2025 and has filed for eighty-eight thousand satellites. Google is aiming, under Project Suncatcher, at eighty-one satellites within one kilometre, linked by laser and fitted with its own processing chips. SpaceX and xAI stand on the American regulator's docket with one million compute satellites.

Distribution across multiple sites is therefore under way. It is under way for compute.

The rationale is sober and has nothing to do with survival. Orbit offers sunlight without night, no neighbourhood objections and no cooling water. Space is being opened up as building land because it is cheaper than a permit on Earth.

This inverts Garrett's picture. He hopes we distribute ourselves before our machines become too fast. In practice the machines distribute first. The backup is being built, and it is being built for compute. We stay where we are.

What Counts When Insurance Is a Proof

Once insurance depends on a threshold rather than an intention, what counts as progress shifts.

A successful landing is then not progress. An announced settlement is not. A schedule with dates attached is not. Progress is a demonstrated unsupported duration, measured against a resupply interval that is itself demonstrated. Everything else is intent.

This is an uncomfortable adjustment, because it devalues almost everything currently celebrated as a milestone. It has one advantage no declaration of intent can offer. It can be checked. An installation can be closed and its endurance measured. Loops for air, water and food can be documented with numbers. Spare-part demand can be plotted over time until the curve crosses supply.

And this is exactly where the question begins that no engineer answers alone. A proof needs someone to accept it. It needs a procedure by which it is examined. It needs a body able to object when a figure does not hold. For aircraft, nuclear installations and pharmaceuticals we have had this for decades. For the place where humanity is supposedly meant to outlast its own failure, we do not.

The Deadline That Keeps Running

I consider Garrett's number more precise than it can afford to be. One to two hundred years follows from assumptions nobody can verify. What remains from his paper is sturdier than the number. It is the ratio between two speeds, and that ratio holds.

The deadline is therefore not the real problem. The problem is that we have no measure by which to read whether we are meeting it. Without a threshold every landing counts as a step toward safety, even when it carries us not one metre in that direction. Without a threshold, insurance can be claimed for as long as the money lasts.

A fallback site is not an escape and not an alternative to Earth. It is the proof that we can operate an environment that was not given to us. As long as that proof is outstanding, we have exactly one place. What we build beside it are appendages that share our fate, merely with a delay.

Insurance is a proof, not a declaration of intent. Who could produce that proof today for a site beyond Earth, and who would accept it?

Website: https://planet-futures.org