Nobody agreed on when The Aryndis Experiment actually started.

Officially, it began with a launch window and a press release about “closed-loop resilience studies in isolated environments.” There was a logo. A slogan. A rotating 3D render of something that might have been a station, or a habitat, or just an artist’s optimistic guess. The funding brief used words like sustainability, redundancy, and future-facing human presence. The numbers attached to it were the kind that make oversight committees develop sudden migraines.

Unofficially, it started earlier — when someone high up realized that certain questions were easier to ask if the answers didn’t have to come back.

Aryndis wasn’t on most maps. It was “provisionally designated,” which meant telescopes had seen something, trajectory models had nodded politely, and nobody had bothered to go argue with the math in person. Far enough to be inconvenient. Close enough to pretend it wasn’t exile.

The crew manifest read like a compromise between disciplines and deniability. Engineers who signed long NDAs. A few scientists whose previous projects had ended in phrases like data irregularities. Medical staff trained more in observation than intervention. There was even a sociologist, which should have made people nervous, but the budget line just said behavioral metrics.

Everything about it was overbuilt. Triple-redundant life support. Radiation shielding thick enough to survive a moderate apocalypse. Autonomous fabrication bays. If something broke, Aryndis could print a new one. If someone broke… well. There were protocols.

The public explanation was simple:
“Assessing operational integrity under extended isolation.”

The private addendum, the one not stored on any network, was shorter:
“Documenting emergent trends within isolated modules.”

For the first few months, reports were immaculate. Air composition stable. Crop yield within projections. Power surplus. The habitat logs were so clean they felt staged, like everyone had agreed to perform competence for an invisible audience.

Then the language shifted.

Not in alarms or failure notices — those would have triggered intervention, and intervention was explicitly what the design avoided. The shift was subtler. More qualifiers. More references to “emergent conditions.” The word baseline started appearing a lot, usually next to phrases like no longer applicable.

Someone on Earth tried to schedule a live Q&A with the crew. It was quietly canceled. “Bandwidth allocation issue.”

The funding increased the following quarter.

By the second year, supply ships were still being sent, but their mission profiles changed. Less cargo, more sealed containers. The return manifests grew thinner. Samples were transmitted digitally instead. “Material integrity concerns,” the memo said, which could mean anything from dust contamination to existential dread.

A junior analyst — the kind who still believed acronyms corresponded to reality — once asked what the actual success criteria were.

The answer, after a long pause, was:
“Continuity of data.”

There were rumors of a final report. Thousands of pages. Most of it redacted, the rest written in the passive voice, where nothing does anything — it merely occurs. Somewhere in the middle, there’s said to be a paragraph noting that the system at Aryndis had reached a “stable, self-consistent state.”

No one clarified whether that referred to the habitat.

Or the people.

Funding for The Aryndis Experiment was renewed twice after that. Not because it was working.

Because it was interesting.