Console
What is moving.
The live globe, OBSERVED-first. Every feed fused into one picture, updating as it happens, with the provenance chip on every contact.
01The research goal
Nocturn shows what is moving in the world right now and can defend every dot it draws. It is the lab's live answer to a research question: can a real time intelligence picture stay accurate about what it actually observed?
Aircraft, vessels, satellites, seismic events, fire and conflict signals come in from free public feeds, get normalized into one track picture, and land on a single globe. Most of that data is public already. What did not exist was one place to interrogate it, rewind it, and cite it.
It is the first system from the lab to reach public deployment, and it is still shipping.
02How it works
Every observation carries a provenance label and the system never hides the difference between them. OBSERVED means a sensor saw it. INFERRED means Nocturn worked it out and says so. TRACKED and DISPUTED mean what they say. Where a typical tool prints “jamming detected,” Nocturn prints “jam? inferred” and shows the reasoning: no receiver reported jamming, the conclusion was derived, and the confidence is on screen.
That is the provenance rule, and it is enforced in the contracts layer rather than left to good intentions. Zero fabricated positions is a constraint of the system, not a marketing claim.
Live and replay share one render path, so any moment can be rewound and exported with provenance still attached, “show me this airport between 03:00 and 05:00 UTC on the 11th” is a question the tool can actually answer.
The surfaces
4 instruments, one discipline
Casefile
What it means.
Where an analyst builds a sourced record. The objects here are claims, sources, and threads rather than contacts: dossiers accumulate evidence over time instead of being assembled per story, and every claim keeps the source it came from.
Range
What happened.
Authored episodes, a reconstructed scene you can scrub through, badged frame by frame with what was measured versus what was reconstructed. Built for showing someone else what occurred without overstating what was known.
Sandbox
What if. In development.
The modeling surface, not on the console today. Place emitters, jam cones, GPS-denial domes, radar coverage and sensor footprints by hand and watch the interactions. Everything authored here will be MODELED by construction and labeled as such, so a simulation can never be mistaken for an observation.
In development
The journalism programme
The next phase turns Nocturn from an operator's instrument into a tool newsrooms and independent investigators can use.
Existing tooling splits the investigative workflow into pieces that do not talk to each other. Atlos and Aleph handle evidence you have already collected. Datawrapper and Flourish handle stories you have already written. Dataminr owns “what is happening now”, and gives you an alert, not a map you can interrogate, rewind, and cite. It is also closed and enterprise-priced.
Nocturn sits in the gap: live, geospatial, rewindable, and provenance-tracked, in a form nobody currently owns openly. The replay engine is the differentiator, and the provenance rule is the antidote to the AI-slop problem newsrooms are about to have.
One more requirement shapes the whole design: it should not only carry bad news. Dossiers track victories won and action taken, not just harm done.
01
Dossiers on actors
If it is a noun, it gets a dossier: a person, a congressman, an event, an action, a law or resolution. Dossiers accumulate evidence over time rather than being assembled per story.
02
Threads and zones
Circle a zone, open a thread, and have it map globally, so a pattern of related events becomes a single traceable object instead of scattered bookmarks.
03
Report ingestion
Pull data out of reporting that already exists, YouTube, Bloomberg, wire copy, and map it, so prior journalism becomes structured evidence instead of a link in a document.
04
Economic data tracking
Take economic figures already reported in the news, map them, and let related threads connect automatically.
05
Knowledge systems and networks
Formal sensing is not the only way to know something. When indigenous leaders found the children in the Amazon, that was knowledge the satellites did not have. The system should be able to hold that kind of knowledge alongside the instrumented kind.
06
Networks and due diligence
A private surface for audits and background work on corporations, people, communities and towns, including the referral chain itself, where one source names the next, and the pattern of who points where is its own evidence.
07
Field methods
Transcripts, clippings, interviews, surveys, ethnography, and records of barriers and access, the qualitative material that carries an investigation, given the same first-class treatment as a sensor feed.
What the lab built
- Built the shared type and policy contract layer the whole system is written against
- Built the country-focus subsystem and automated daily reporting across ~240 countries
- Built 3D map-rendering features on CesiumJS, including contact and track visualization
- Automated situation reports, delivered where the audience already is
- Every door locked: no part of the system is reachable without a login
- 20,000+ lines contributed by 42 Quantum alongside the project creator
- Lines of code
- 241,850Lines of code
- Commits
- 928Commits
- Files
- 939Files
- Building since
- Jun 2026Building since
- TypeScript
- React
- CesiumJS
- Node.js
- DuckDB
- Public data feeds
At a glance
How the system flows
- 01
Ingest
Aircraft, vessels, satellites, seismic, fire, and conflict signals from free public feeds
- 02
Normalize
One track picture, provenance on every observation
- 03
Correlate
Events linked across feeds
- 04
Render
The live 3D globe
- 05
Replay
Any historical window, frame for frame