orangekraken◉radio
Skip to content

Help OrangeKraken Radio listen to the world

While your node is running, your computer helps measure and label radio signals — a slice of our receiver's spectrum, or the spectrum shape of a sample from a world receiver. No account. No installation. It stops the moment you close it.

Checking the ledger store…
Real work available now: 1352 spectrum slices from today's captures, plus world-receiver spectrum shapes whenever the receiver scanner has fresh ones.

How it works

  1. Your browser asks for one small, real job: look at a slice of radio spectrum, count its peaks, or say what kind of signal a world receiver's spectrum shape looks like.
  2. For spectrum slices, the machine makes a guess (“that's a station”, “that's just noise”) and you confirm it or correct it with one click. That click is the valuable part.
  3. For the judgement tasks, the same job goes to other visitors who don't know you. When at least two independent sessions agree, the answer becomes a verified example. Disagreements are kept on purpose — they are where a model learns the most.

Privacy, exactly

  • No account, no cookie, no fingerprint, no IP is kept — ever.
  • Your session id lives only in this browser tab (sessionStorage) and vanishes when the tab closes.
  • The server hashes that id with a daily salt — just enough to tell “two different sessions” from “one session twice”. The hash never enters an example and is never exported.
  • Sharing results and sharing performance stats are two separate opt-ins. Untick both and the work still runs — entirely on your machine, nothing sent.

What lands in the OKLM training ledger

OKLM is the small language model trained only on this engine's own record. What your node contributes to it is a de-identified row: the task, a reference to the real input (which spectrum cell at which capture time, or which spectrum shape), the final answer, whether humans confirmed or corrected the machine's proposal, and how many independent sessions agreed. Only human-labelled VERIFIED rows are eligible for training; machine integrity checks (such as peak counts) are counted but never trained on. The ledger below is the live, public count of all of it — including the disagreements.

Ledger status
checking…
What is retained (only with the first opt-in):
  • the task, a reference to the real input (cell + capture time, or the spectrum-shape id), and your answer
  • whether you confirmed or corrected the machine proposal
  • how many independent sessions agreed — and the disagreements, kept on purpose
Never: an account, a cookie, a fingerprint, an IP, or the session id (it hashes with a daily salt for the ≥2-sessions check and never enters an example). Privacy