Instruments
One sheet per system. Every one of them is a prototype still under development, and every one of them is running live; the numbers are read from the running system, not from a plan. What it does, since when, what it is wired to, and the number with its measurement date.
What's running
As of 4 November 2026Perception
- RetinaReads more than 200 news sources a week, from newspapers and agencies to public channels, and finds the storylines, the tone and the spikes around people, places and organisations.since March 20261,096,777articles, 4 Nov
- ChimeraA graph of actors with profiles and the links between them, tied to the news. Its own graph engine.since April 202622,146actors, 4 Nov
- GeospatialThe sensor under the situation picture, and its anomaly detection: nine live feeds, four engines, two live graphs that forget.since March 202614anomaly cells active, 4 Nov
- Situation PictureRegional situation pictures, threat levels, dossiers on actors and assets, flights and ships.since March 202662,926objects in 12 classes, 4 Nov
- ObservatoryOpen sources, read at scale, and the site where the picture goes public: news, public Telegram channels, transponders, infrastructure, registers, sanctions lists.since May 202612,706news reports on the public News Deck, 4 Nov
Knowledge and research
- Knowledge GraphBooks, papers, statutes, memory and code in one memory core. Ask a question in plain language, get answers from several fields at once.since March 2026334,670text passages, 8 Oct
- Findings EngineDid research on its own at night until 25 September 2026: collected findings across fields, interrogated them, spotted gaps. Rebuilt on the new core since 8 October, with 502 findings carried over.since 6 June 2026524findings in the archive, 8 Oct
- Formula PipelineMade formulas from textbooks structurally comparable across disciplines, with its own parser. Stopped in July 2026; the formulas are kept as an archive.since July 2026274,244formulas, 8 Oct
- Semantic Research EngineFolded the symbols of different disciplines onto shared concepts and went looking for the definitions it was missing. Stopped on 1 August 2026; what it built is archived.since 3 March 202681,708symbols folded onto 12, 8 Oct
Why the lab watches, and where it stops
The perception instruments read like intelligence tooling, and a visitor is right to ask what they are for. The purpose is a test: whether one human and one agent can build a perception layer out of open sources that holds up against what actually happens, and whether a machine can be made to say how sure it is and where it was wrong. The situation picture is the surface of that test. It is read by one person, the person who built it. It has no customers, no subscribers, and nothing on it is sold or offered as a service.
The limits are the ones the sources set. Everything comes from what is openly published: the press, public channels, transponders that broadcast in the clear, registers and sanctions lists that any citizen may consult, satellite fire detections that a space agency puts online for everyone. Nothing is scraped from behind a login, nothing is bought, and the original articles stay with their publishers; the lab keeps what it derived from them. People appear only as they appear in the press, as public actors in public events. The lab does not track individuals, does not build profiles of private persons, and does not take tasking from anyone. The map covers seventeen conflict zones, not the world, because the question is whether the reading holds, not how much can be seen.
What that leaves is a lab reading the same news everyone reads, with better bookkeeping, and saying out loud when its own reading failed. The pages on Retina, Chimera and the situation picture each state what is not yet there. That is the part of this work the lab considers most worth publishing.
How the lab does research, and what it still owes
The instruments under Knowledge and research are not a toolbox, they are one programme at different stages of completion. The idea behind it is that a claim should have to survive a route rather than merely be recorded: state it, check whether someone has already published it, establish where it breaks, look for a counterexample, derive a prediction from it, and only then treat it as a finding. Each of those is a station with a gate, and each gate is a method in its own right that has to be built, tested against a case where it must fail, and then switched on.
Two stations are running. Findings are recorded with the passages they stand on and linked to each other, and since 2 September 2026 a new one is checked against the open record of published literature before it is accepted, provided the engine is confident enough in it to be worth the check. As of 3 September 2026 the graph holds 454 findings, tied together by 211 supporting and 43 contradicting links. That second number matters more than the first: contradictions are kept and turned into questions instead of being deleted, and a body of findings with no contradictions in it would not be a research result, it would be a collection.
The rest is owed. The counterexample stage, the deliberate self-experiment, and the calibration test in which the machine is asked to rediscover a known result without being told it exists are designed and not yet built. The programme is written down as a numbered plan, and once a week an automated run puts it back in front of its author and counts only the items that come with evidence, which is why the honest figure for progress is usually lower than the one the machine reports about itself.