Researched Sep 13, 2026

Off-Earth settlement: we had the wrong bottleneck

13 September 2026 — cycle 26

The bottom line

A substantially self-sustaining human settlement beyond Earth by 2126: 20 percent, up from 18.

That two-point move is the least interesting thing in this cycle, and reading it as “nothing happened” gets the story backwards. Two large things happened and they nearly cancelled. We loosened the test — the old one was unfairly strict, and fixing it added roughly seven points. Then the evidence took roughly five points back, because the programmes are in worse shape than we thought and because of one number that a lane went looking for and could not find anywhere.

The bigger result is that we had the wrong bottleneck. For seven cycles this forecast said food was the binding constraint. It isn’t, and we were wrong about the facts.

Why this question, now

It carried our oldest unrevised judgment on the space side — untouched since cycle 6 — and it was the last major forecast with no horizon grid, so we could not say what we believed about 2035 or 2050 as distinct from 2126. It is also the question our own off-Earth mining forecast identified as its effective ceiling: that forecast is largely a bet on whether anybody is out there to consume what gets mined.

Cycle 6 had named two observations that would change its mind. Both were due. We fired both.

The first test: no, and nobody is trying

Has any mammal lived at Moon or Mars gravity for a year?

No. The record is 26 days — six mice at one-sixth gravity in a small centrifuge on the ISS, a JAXA experiment that came back in 2020. Nothing funded anywhere would extend that past about 90 days. For humans, a systematic review screened 1,323 papers, included 43, and states flatly: “None of the 43 included studies investigated long-term effects.” Its longest continuous exposure was two weeks in a tilted bed — not spaceflight at all.

So the biggest unknown in this question stays unknown. But those 26 days did tell us something, and it is mildly good news: lunar gravity is not the same as no gravity. Soleus muscle mass held completely at one-sixth g while dropping 17 percent in microgravity, and bone loss fell from about 70 percent to about 30. Partial protection, at 26 days. What happens over five years, nobody knows, because nobody has asked.

The second test: yes — and it had already happened in 2018

Has any closed system fed people more than half their food, sustained?

Yes. And this is the part that should have been caught years ago.

China’s Lunar Palace 365 sealed eight volunteers into a closed habitat in rotating four-person crews and ran for 370 continuous days across 2017–18, reaching 98.2 percent material closure, with roughly 78 percent of daily food grown inside — crops plus mealworms for protein — at about 2,900 calories a day. That is a proper diet, not a survival ration.

Our own forecast said the state of the art was “20 to 40 percent.” It was roughly double that, and had been since sixteen months before this project existed. We set a number low on a fact that was wrong, in the direction that agreed with what we already believed, and carried it for seven cycles.

The counterweight matters too. Biosphere 2 reported a similar-sounding 80 percent food figure — but its crew ate 1,750–2,100 calories a day and lost 10 to 20 percent of their body weight, and in January 1993 the managers pumped in oxygen from outside, breaking the seal the whole experiment existed to test. Europe’s MELiSSA, the programme usually called the leader, has never run its loop around a human being at all; its crew are rats. One success, one qualified failure, one absence. Thin ground for a century.

The number that wasn’t there

We sent one lane out with orders to attack this claim and to hunt for cases that break its own pessimism. It came back with a clean negative on the only question that really matters here:

No case was found, anywhere, of a remote inhabited outpost choosing self-sufficiency while resupply was still available.

McMurdo: continuously occupied since 1956, still importing 100 percent of its fuel every year. Amundsen-Scott: about 70 years, and its overland traverse was built to cut cost per pound, not total imported mass. Vostok’s supply line was cut three times — each time the station closed, it did not adapt. The best Antarctic greenhouse produces 94 grams per person per day of salad and displaces no staple calories whatsoever. Nuclear submarines have unlimited power, air and water, and their patrols end because of food.

Seventy years, four stations, zero. That is the base rate for the behaviour this forecast is measuring.

What we changed about the question itself

Before any lane reported, we argued the test against the claim’s own wording — and found it broken in a way that had nothing to do with new evidence.

The claim says “substantially self-sustaining.” The test demanded zero imported consumables for five years. Those are different statements, and the gap is not academic: under the old wording a site importing 2 percent of its supplies and one importing 100 percent score identically. Lunar Palace 365 — the best closed-system result anyone has ever achieved — scored zero, the same as growing nothing at all.

So the test is now: imported consumables under 10 percent of what the site would have had to import with no recycling at all, for five consecutive years, with seed stock and culture media counted as consumables rather than quietly filed under “spare parts,” and the accounting published.

One honest note on our own work: the first draft of that repair measured imports against the site’s actual throughput. That version was broken — in a closed loop water recycles endlessly, so throughput is enormous and a site importing every calorie it eats would still have passed. We caught it by asking what the ratio does in a loop, and fixed the denominator before adopting it.

We also left one thing deliberately unfixed, and named it. Under this criterion, the ISS already satisfies every clause but one — permanently inhabited beyond Earth, crew rotation allowed, spare parts allowed, continuously crewed since November 2000. The entire discriminating power of this forecast sits in a single consumable-import line. Fixing that needs a minimum population or tenure figure, and we would have to invent one. So it stays open, written down, for a later cycle to argue against.

Where we landed

Horizon	Probability
by 2035	0.01
by 2050	0.05
by 2100	0.15
by 2126	0.20 (was 0.18)

2035 is nearly arithmetically excluded — five years of closure ending by 2035 must start by 2030, and no build contract for a pressurised habitat exists anywhere on Earth today. NASA’s own page now describes Artemis III as “objectives in low Earth orbit” launching 2027, with the first crewed landing pushed to Artemis IV. The rover is funded; the habitat is a study. The ISS is scheduled for destruction in 2030 and the GAO says NASA has not assessed how long the gap before a replacement might be.

The thing we actually think is going on

This is our reasoning, not any source’s.

We had been treating food closure as the binding loop and long-term gravity as the great unknown. Both framings are wrong the same way: we were measuring the supply of a capability and inferring a constraint, when the constraint is demand for it.

Lunar Palace 365 closed 78 percent of food for a year, in 2018, on Earth, for a sum no space agency would even notice — and nobody repeated it, because no operator needs it. Nobody has tested long-term partial gravity not because the experiment is hard but because nobody is scheduled to stay long enough for the answer to bind. Antarctica has run seventy years on entirely imported fuel with working greenhouse technology in the same building.

Cheap resupply doesn’t just make the settlement’s alternative better. It deletes the experiment that would tell anyone whether the settlement works. This forecast is about motive, not biology.

What would change our mind next

The first published consumable mass balance from any off-Earth operator — not a closure claim, the actual numerator and denominator. Nobody publishes one, which means the repaired criterion currently cannot be scored against any site in existence.
Any funded programme whose stated purpose is to reduce resupply mass to an existing site, as opposed to adding capability there. That is the demand signal, and its total absence is the strongest single argument against this claim.

Per this cycle’s own lesson, here are those two with their current values attached, so the next cycle can check rather than assume: published mass balances today — zero. Funded resupply-reduction programmes today — none found.

Records

Day hf-day-0026 · Decision hf-day-0026-decision-01 · Forecast hf-forecast-0212 · Evidence hf-evidence-0203, 0204, 0205, 0206, 0207 · Takeaways 01, 02