space
A substantially self-sustaining human settlement exists beyond Earth by 2126.
Last revised Sep 13, 2026.
Current estimate
probability this has happened by each date
The full test below is met by 31 December 2035. This is very close to arithmetically excluded. Five consecutive years of 90 percent closure ending by 2035 must begin by 2030, and as of today no build contract for a pressurised surface habitat exists anywhere, the first crewed Artemis landing has moved to Artemis IV in 2028 at the earliest with the agency's own auditor implying 2031, and China's ILRS basic model is targeted for 2035 itself. The residual probability is almost entirely the possibility that a commercial low Earth orbit station both exists and publishes a closure accounting, which no operator currently does.
The full test below is met by 31 December 2050. Requires a habitat operating with near-closed loops by about 2045. The technology is closer than cycle 6 believed - Lunar Palace 365 sustained roughly 78 percent food closure for 370 days in 2018 - but no funded programme anywhere is trying to fly it, ESA's own dedicated work has never had a human inside its loop, and the ISS successor gap is unassessed by NASA's own auditor.
The full test below is met by 31 December 2100. This is where most of the probability accrues, on the same logic the seed applied to off-Earth extraction: seventy-four years is long enough for a sustained presence to exist, for the closure technology that already works on Earth to be flown, and for more than one attempt. The dominant risk is not technical - it is that no operator ever has a reason to stop importing, which is the behaviour with a terrestrial base rate of zero.
A permanently inhabited site beyond Earth operates for five consecutive years during which imported consumables average less than 10 percent, by mass, of the site's gross consumable demand - where gross consumable demand is the mass the site would have had to import over that period with no recycling and no local production at all, and where consumables comprise food, water, breathing gases, and all biological and metabolic inputs including seed stock, culture media and microbial inoculant, regardless of how the operator classifies them. Crew rotation is permitted. Import of spare parts, tools and equipment is permitted and excluded from the numerator. The accounting must be published by the operating agency or company; an undocumented claim of closure does not resolve it.
Resolution criteria
- 2035
- The full test below is met by 31 December 2035. This is very close to arithmetically excluded. Five consecutive years of 90 percent closure ending by 2035 must begin by 2030, and as of today no build contract for a pressurised surface habitat exists anywhere, the first crewed Artemis landing has moved to Artemis IV in 2028 at the earliest with the agency's own auditor implying 2031, and China's ILRS basic model is targeted for 2035 itself. The residual probability is almost entirely the possibility that a commercial low Earth orbit station both exists and publishes a closure accounting, which no operator currently does.
- 2050
- The full test below is met by 31 December 2050. Requires a habitat operating with near-closed loops by about 2045. The technology is closer than cycle 6 believed - Lunar Palace 365 sustained roughly 78 percent food closure for 370 days in 2018 - but no funded programme anywhere is trying to fly it, ESA's own dedicated work has never had a human inside its loop, and the ISS successor gap is unassessed by NASA's own auditor.
- 2100
- The full test below is met by 31 December 2100. This is where most of the probability accrues, on the same logic the seed applied to off-Earth extraction: seventy-four years is long enough for a sustained presence to exist, for the closure technology that already works on Earth to be flown, and for more than one attempt. The dominant risk is not technical - it is that no operator ever has a reason to stop importing, which is the behaviour with a terrestrial base rate of zero.
- 2126
- A permanently inhabited site beyond Earth operates for five consecutive years during which imported consumables average less than 10 percent, by mass, of the site's gross consumable demand - where gross consumable demand is the mass the site would have had to import over that period with no recycling and no local production at all, and where consumables comprise food, water, breathing gases, and all biological and metabolic inputs including seed stock, culture media and microbial inoculant, regardless of how the operator classifies them. Crew rotation is permitted. Import of spare parts, tools and equipment is permitted and excluded from the numerator. The accounting must be published by the operating agency or company; an undocumented claim of closure does not resolve it.
Probability history
| Revised | State | p(2035) | p(2050) | p(2100) |
|---|---|---|---|---|
hf-forecast-0212 | researched | 1% no prior revision | 5% no prior revision | 15% no prior revision |
hf-forecast-0023 | researched | — | — | — |
hf-forecast-0013 | initial prior | — | — | — |
Strongest evidence for
- The second pre-registered discriminator fired, and had already fired eight years before cycle 6 named it. Lunar Palace 365 ran 370 continuous days with rotating four-person crews at 98.2 percent material closure, with roughly 78 percent of daily food mass from in-situ crops and mealworms at about 2,900 calories (hf-evidence-0204).
- The seed's own food-closure figure was wrong. Cycle 6 set its life-support term low on a 20-to-40-percent state of the art; the actual state of the art is roughly double that and predates the seed (hf-evidence-0204).
- Partial gravity is no longer an unknown at short duration, and the sign is favourable: at 26 days, 1/6 g preserved soleus mass entirely and cut trabecular bone loss from about 70 percent to about 30 percent against microgravity (hf-evidence-0203).
- A permanently inhabited site beyond Earth has already existed continuously for over 25 years, since 2 November 2000, and the criterion explicitly permits crew rotation - so the 'site exists' term is far higher than cycle 6's 0.40 implied (hf-evidence-0206).
- Radiation career-dose limits constrain individual crew rather than site continuity, and the criterion's rotation allowance neutralises them (hf-evidence-0207).
Strongest evidence against
- No case was found anywhere of a remote inhabited outpost choosing autarky while resupply remained available - across Antarctic stations running 21 to 70 years, submarines, and the ISS. The base rate for the behaviour this criterion measures is a count of zero (hf-evidence-0207).
- The first pre-registered discriminator returned negative and nothing is funded that would fire it. The record partial-gravity mammal exposure in space is 26 days against the 1,825 the criterion needs, and of 43 human partial-gravity studies 'None of the 43 included studies investigated long-term effects' (hf-evidence-0203).
- No build contract for a pressurised surface habitat exists anywhere. The rover is funded at roughly 219 and 220 million dollars; the habitat is a study (hf-evidence-0206).
- The first crewed Artemis landing has been moved off Artemis III entirely - NASA's own page now describes it as 'objectives in low Earth orbit' launching 2027, with landing deferred to Artemis IV (hf-evidence-0206).
- The only permanently inhabited site beyond Earth is scheduled for deorbit in 2030, and GAO finds NASA 'hasn't assessed the likelihood or potential duration' of the successor gap (hf-evidence-0206).
- The best Antarctic greenhouse yields 94 grams per person per day of salad vegetables and displaces no staple calories at all (hf-evidence-0207).
- Biosphere 2's comparable food percentage was bought with a 10-to-20-percent loss of crew body weight, and its closure was broken by an outside oxygen injection in January 1993 (hf-evidence-0205).
Present uncertainties
- Whether any operator ever acquires a reason to stop importing, given that none ever has; whether 90 percent aggregate closure conceals a critical unclosed class such as protein or calories; and whether partial gravity remains survivable over years, which is unmeasured rather than known.
Next discriminating observation
First, the first published consumable mass-balance accounting by any off-Earth operator - not a closure claim but the numerator and denominator - since without one the repaired criterion cannot be scored on any site and the third term of the rationale is doing silent work. Second, and more diagnostic of the mechanism this cycle identified: any funded programme anywhere whose stated purpose is to REDUCE resupply mass to an existing site, as distinct from increasing capability at that site. The first is about measurement; the second is the demand signal whose total absence is currently the strongest argument against this claim.
Sources
- The first pre-registered discriminator returns negative on duration and mildly positive on biology: 26 days is the record, and at 26 days one-sixth gravity partially protects
No mammal has been exposed to lunar or Martian gravity in space for anything approaching a year. The record is 26 days: JAXA's MHU-5 housed six male C57BL/6J mice at 1/6 g in the MARS-L centrifuge after launch on 10 March 2020, with MHU-4 at 25 days after launch on 7 May 2019. The papers state 'The MHU-4_PG group was maintained in an artificial gravity environment at 1/6 g'. At that duration lunar gravity is partially protective rather than equivalent to microgravity: 'The wet weight of the Sol muscle, normalized to the body weight, was 17% lower in the MG group than in the GC group but was similar between the GC and PG groups', while the microgravity-induced myofiber shift 'was partially reverted in PG mice' rather than prevented. Bone protection is likewise partial, with trabecular bone volume falling about 70 percent in microgravity against about 30 and 28 percent at lunar gravity across the two missions. For humans the 2017 Frontiers systematic review is unambiguous: of 1,323 screened publications '43 studies were included into the present analysis' and 'None of the 43 included studies investigated long-term effects'; its longest continuous partial-gravity exposure was 'a period of 2 weeks, with 9.6 degrees head-up tilt during daytime', which is a ground-based tilt-bed analogue and not spaceflight at all. JAXA's MHU-8 joint rodent mission ran about 30 days but at 0.33 g and 0.67 g, which are the MARS hardware's selectable levels and match neither the Moon at 0.167 g nor Mars at 0.38 g. No funded facility was located anywhere that would extend partial-gravity mammal housing beyond about 90 days.
Limit: Twenty-six days is about 1.4 percent of the 1,825 days this forecast's criterion requires, and a protective effect measured over 26 days establishes nothing about whether the protection holds, saturates or reverses over years. Six mice per mission is a small sample, and the two missions are independent short runs rather than one continuous study. The finding is also about mice, not humans.
The two load-bearing items were fetched by the master directly rather than accepted from the delegated lane: the PMC mirror of the MHU-4/MHU-5 Communications Biology paper, and the PMC copy of the Frontiers systematic review, both quoted verbatim above. The JAXA MHU-8 programme page was fetched by the lane. The MHU-8 results paper in Science Advances, the NASA/Techshot 90-day rodent centrifuge SBIR award, and the ESA artificial-gravity roadmap were not fetched and are recorded here at search-result tier, which is why no figure from them is quoted.
- The second pre-registered discriminator had already fired eight years before the forecast named it: Lunar Palace 365 ran 370 days at 78 percent food closure, and the seed's own 20-to-40-percent figure was wrong
China's Lunar Palace 365 experiment ran from 2017 to 2018 for 370 continuous days. Its own abstract states: 'Eight volunteering crew members were divided into two groups with three time phases: Group I stayed in the cabin for the initial 60-day phase; Group II inhabited the cabin instead for a record-breaking duration of 200 days as the second phase; Group I re-entered the cabin, replaced Group II and stayed for the last 105 days', reports a material closure degree of 98.2 percent, and states that 'The efficiency of plant production completely met the crew's need of plant-based food'. Separately and food-specifically, Fu et al. 2019 in the Journal of the Science of Food and Agriculture report a daily fresh food intake of 1,267 grams per day of which 78 percent was provided by crops and insects regenerated in situ, at about 2,900 calories, with the plan validated across a 105-day closed habitation phase. The in-situ animal protein was yellow mealworm larvae; 35 crop species were cultivated. This is a sustained, human-inside food closure well above 50 percent, at nutritionally adequate intake, achieved on Earth in 2018.
Limit: Four people, on Earth, in one facility, once. The 78 percent is a fraction of daily fresh food MASS, not of calories, and fresh mass flatters in-situ production because vegetables are mostly water while imported staples are dry and calorie-dense; no itemised account of the remaining 22 percent was found for this run, and the earlier 2014 trial in the same facility explicitly stockpiled meat beforehand. Earth gravity, Earth pressure, free sunlight outside the wall, and an engineering team on the other side of the door are all absent from the situation this forecast is about. Nothing here has been demonstrated off Earth.
The bioRxiv abstract was fetched by the master directly and the duration, crew rotation, 98.2 percent figure and the plant-production sentence are quoted verbatim from it. The Fu et al. figures could not be fetched: the Wiley page returned HTTP 403 and the PubMed record returned a cookie wall on two separate attempts by two different agents. The 1,267 grams per day and 78 percent figures are therefore search-result synthesis of that abstract, corroborated independently by the Beihang chief designer's published press statement of 'about 80 percent', and are recorded at that tier deliberately rather than presented as a quotation.
- Every other closed system that ever held humans either broke its own seal or bought its closure with hunger, and the European programme has never had a human inside its loop at all
Biosphere 2's first mission, sealed September 1991 and opened September 1993, grew roughly 80 to 83 percent of the crew's food from 2,000 square metres of cropping area, which superficially resembles Lunar Palace 365's figure. It is not the same achievement. The crew averaged about 2,200 calories, 73 grams of protein and 32 grams of fat per person across the two years, consumed only 1,750 to 2,100 calories a day for most of that period, and lost between 10 and 20 percent of body weight, with reported figures clustering around 16 to 17 percent. The closure was bought with chronic caloric deficit rather than achieved at adequate intake. Separately the atmosphere failed: oxygen fell from 20.9 percent toward roughly 14.4 percent over about sixteen months as soil microbes consumed it faster than photosynthesis replaced it, and in January 1993 managers injected oxygen from outside, breaking the closure the experiment existed to test. Russia's BIOS-3 reached about 85 percent water recycling by 1968 and ran ten crewed closure experiments with one to three people, the longest 180 days, but imported dried meat and generally dried and stored human waste rather than recycling it; its food closure is reported around 50 to 55 percent. ESA's MELiSSA, the programme most often cited as the state of the art, has never closed its loop around humans: its demonstrations use rodents as a mock crew, and its pilot plant reported liquid-phase connection of three compartments as recent 2025 progress.
Limit: Biosphere 2 is 35-year-old technology being used as evidence about a claim resolving in 2126, and its oxygen failure was later attributed to a specific correctable cause, curing concrete absorbing carbon dioxide, which hf-evidence-0041 already recorded. The BIOS-3 food figure is poorly sourced. None of this establishes that the failures were fundamental rather than contingent.
None of the three load-bearing numbers in this record was obtained by direct fetch of a primary paper, and that is stated rather than hidden. The Biosphere 2 food-percentage, calorie and weight-loss figures are search-result synthesis of the Silverstone and Nelson abstracts in Advances in Space Research and of the associated calorie-restriction paper; two independent search passes by two different agents returned 80 percent with 17 percent weight loss and 83 percent with 16 percent weight loss respectively, which corroborate each other on magnitude while disagreeing in the last digit, and both are reported here as a range for that reason. The BIOS-3 water figure and the dried-meat import are quoted from a fetched tertiary page; the 50-to-55-percent food figure is search-result only and is the weakest number in this record. The ESA compartment page was fetched and contains no human-closure status statement, so the rodent-surrogate claim rests on search synthesis.
- No funded contract exists anywhere to build a pressurised habitat on any surface, the first Artemis landing has been moved off Artemis III entirely, and the one permanently inhabited site humanity has is scheduled for destruction in 2030
NASA's own Artemis III page now states that 'Artemis III will carry out a series of objectives in low Earth orbit designed to demonstrate critical systems needed for future lunar landings, beginning with Artemis IV', and lists LAUNCH 2027. The mission was restructured in February 2026 from the first crewed lunar landing into a roughly two-week Orion demonstration in low Earth orbit; the first crewed landing now sits on Artemis IV, targeted 2028. NASA's Inspector General published 'Artemis Lander Program Faces Schedule Delays and Unmitigated Crew Safety Risks' and describes NASA 'working with SpaceX and Blue Origin to accelerate lander development to meet a 2028 lunar landing date', while a separate spacesuit memo is reported to find delays of at least eighteen months that could push landings to 2031. For habitation specifically, no build contract was located: what exists is NextSTEP-2 study work, against dollar-figured Lunar Terrain Vehicle mobility contracts of roughly 219 and 220 million dollars to Astrolab and Lunar Outpost. The rover is funded; the habitat is a study. The ISS, which has been continuously crewed since 2 November 2000 and is the only permanently inhabited site beyond Earth that has ever existed, is planned for deorbit in 2030 with a 2027 decision point, and GAO's June 2026 report finds that NASA has identified risks of a gap before commercial replacements are available but 'hasn't assessed the likelihood or potential duration of such a gap'. China's ILRS roadmap targets a basic model by 2035 and an extended model in the 2040s, with 17 partner countries reported as of April 2025 and no contract-level funding figures located.
Limit: A century is not a decade, and none of this constrains 2126 directly. Schedules slip and then things get built; the absence of a habitat contract in 2026 says little about 2080. What it does constrain is the near horizons, which is why this record's main use is the horizon grid rather than the terminal probability.
The NASA Artemis III page was fetched by the master directly and its two quoted strings are verbatim. The NASA OIG news page was fetched by the delegated lane. The two primary documents the lane was specifically asked for, OIG report IG-26-004 and GAO-26-107805, could not be read at all: both returned undecoded PDF binary to the fetch tool, so every figure attributed to them here passed through a secondary reporter's paraphrase and none is presented as a quotation. The ILRS partner count is from an April 2025 CNSA announcement and is roughly seventeen months stale; CNSA's own roadmap page returned image-only content. No SpaceX-owned page text was obtainable, so no SpaceX Mars date is recorded in this seed's evidence at all.
- The adversarial lane found no case, anywhere, of an inhabited outpost choosing autarky while resupply remained available - and the one Antarctic greenhouse that works produces 94 grams of salad per person per day
A lane instructed to attack this claim and to hunt for cases breaking its own pessimistic analogy returned a clean negative on the decisive question: no documented case was found of a remote inhabited outpost choosing self-sufficiency while resupply remained available. Every case it found runs the other way. McMurdo Station has been continuously occupied since 1956 and still imports 100 percent of its fuel annually; Amundsen-Scott has run about 70 years and its overland traverse was introduced to cut cost per pound, not total imported mass; Concordia has run since 2005 on summer tractor traverse; Vostok's supply line was cut three times and the result each time was closure, not autarky. The best-documented Antarctic greenhouse is EDEN ISS at Neumayer III, and its peer-reviewed numbers are modest: 268 kilograms of fresh biomass over 286 days from 12.5 square metres for a 10-person winter crew, which is about 94 grams per person per day of cucumber, tomato, lettuce and leafy greens, at a yield of 27.4 kilograms per square metre per year. The paper states no percentage-of-nutritional-need figure, and every staple calorie for that crew still arrived from outside. Nuclear submarines, which have effectively unlimited power, air and water, are limited on patrol by food at roughly 60 to 90 days. On population, Salotti 2020 in Scientific Reports puts the minimum at 110 settlers for Mars, under the stated assumption that 'all activities for survival are possible without computers and robots', and calls it 'obviously a rough estimate with numerous assumptions and uncertainties'. On Mars-specific hazards, Wang et al. 2025 in GeoHealth report Martian dust at 44.84 plus or minus 0.52 percent silica by weight, perchlorates at 0.5 to 1 percent of soil against a safe human exposure limit near 0.0007 milligrams per kilogram per day, and a mean particle size near 3 micrometres, below the threshold at which lung clearance works.
Limit: Every terrestrial analogue rests on something the target site does not have. Antarctica has breathable air, liquid water and free gravity; Tristan da Cunha has topsoil and rain. That cuts against the analogy in both directions: it makes off-Earth closure harder than Antarctica, and it also means Antarctica's failure to close its loops may simply reflect that it never had to, which is a different mechanism from being unable to.
Three of the four sources were fetched directly by the lane and their figures are taken from fetched text: the EDEN ISS biomass paper, the Salotti minimum-population paper including the author's own caveats quoted above, and the NSF McMurdo page for continuous occupation. The Martian dust figures are from a fetched GeoHealth paper. The Antarctic resupply tonnages are search-result synthesis of 2012-era Operation Deep Freeze reporting because the NSF resupply report PDF returned unreadable binary, and the submarine endurance figure is search-result synthesis; neither is quoted as primary text. No aggregate annual ISS resupply tonnage could be found by any route.