How the register grades a claim

A record says what was claimed, in the claimant's words, and how good the evidence is. It does not say whether the thing worked.

What counts

A record is a claim that an AI ran beyond the Kármán line, 100 km above the Earth: in orbit, on a space station, in deep space or on another world. AI here means a learned model, or software that plans, perceives or decides on board.

Software that runs on the ground and plans for a spacecraft is left out. Where such a case marks the boundary it is shown beside the register, with no mark.

The mark

A mark belongs to a claim. Its letter says what is claimed to have happened; its number says who says so. A record shows its best mark at each stage.

A Announced. A plan, a filing or a funding round. Nothing above the line.
L Launched. The hardware is reported above the line.
R Ran. A named AI workload is reported to have executed there.
U Used. Its output is reported to have been acted on in operations, repeatedly.

0 Secondary only: news, another tracker, a supplier.
1 The operator's or developer's own statement: a release, a project page, a post.
2 Their own technical account: a paper, a preprint, an abstract with specifics.
3 Confirmed by a party with no stake in the claim, or by public data.

Who can earn which number

The number is set by who is speaking and in what kind of source. It is not the reviewer's opinion of the claim, and a test fails the build if a record's number is higher or lower than its sources carry.

A paper by the mission team earns the mark for a technical account even when it is peer reviewed. Peer review is not a party with no stake.
The operator of a host computer does not confirm a run that it presents as its own work.
A supplier's page, a news report, a directory or another tracker cannot raise a claim above the lowest number, however sure it sounds.
A public satellite catalogue confirms that an object was launched, and when. It cannot confirm that anything ran on it.

Quotations

Every quotation is copied from a saved copy of its source, and a test fails the build if the words are not found there. A shortened quotation is marked with three dots, and each part must be found. Under each quotation the record says who said it, how they stand to the operator, when it was published, when it was read and how much of the source was read.

Where a record sits

Ran. Someone who built or ran the system says an AI workload ran.
Launched, no run reported. The hardware is above the line and no source read reports a run. The silence may be the register's.
Held back. Listed elsewhere as flown, and nothing read says so. Not counted.
Beside the register. A case that marks the boundary. It carries no mark.

Classes

planning Software that makes a plan on board and carries it out.
perception Models that interpret what a sensor takes in: images, spectra, signals.
learning Training a model in space, not only running one trained on the ground.
language Language models, with or without vision.
compute Computers sent up to run AI, whatever has or has not been run on them.
classical Autonomy by fixed algorithms, with no learned model named. In the register, and kept out of the count of models run.

What is automatic and what is not

Jobs run on a schedule. One looks up catalogue numbers. One looks for papers, releases and news that have not been read. One fetches every cited source again to see whether its quotations are still there. Each writes a report for a reviewer, and none can change a record.

A record changes only when a reviewer edits it, the tests pass and the change is merged. Every change is written in the change log, with its reason.

Who reviews

Every record was read, quoted and graded by an AI session on behalf of the owner, on 4 Oct 2026. No person has reviewed them.