The method
We verify what people holding public power say. It's work that touches other people's reputation and the trust of those who read us: that's why the method, even before the software, is an ethical matter.
This summary document sets out the method with which Cryterio isolates and verifies the factual claims made live. It's meant to be read, printed or exported as a PDF. The extended versions of each part are on the Ethical principles, The model and Technology pages.
The method, in four steps
A broadcast's path is linear: we capture the audio of the public stream, we transcribe it, isolate the verifiable claims, compare them against the sources and publish the verdict on the real-time timeline. Each step is an isolated component: this makes the system robust and lets us improve one without touching the others. The broadcasts we verify are chosen by the editorial team under a stated criterion: events of public interest — question time, hearings, rallies, institutional speeches — selected for their news value, not for the political side of the speaker.
The source before the judgement
No verdict without a primary source cited. We never say a claim is false because “everyone knows”: we write it only when a public, verifiable source contradicts the data, and we place that source next to the verdict, with title, body, link and excerpt. If we can't show the evidence, we don't issue the judgement.
When in doubt, caution
When the evidence isn't enough, the verdict is non_verificabile — and that's perfectly fine. A wrong assessment does more damage than a claim left pending. We'd rather say “we don't know yet” than guess. The machine is tuned to be cautious, not to fill the timeline.
We verify facts, not people
Our object is the claim, not who utters it. A Cryterio verdict says “this figure is imprecise compared with source X”, never “this person is lying”. It's a substantive distinction, not one of form: it protects the dignity of the person being checked, reduces the risk of defamation and keeps us anchored to the solid ground of the data.
The same yardstick for everyone
Cryterio has no political colour and does not choose claims based on who makes them. The same data, from the same source, gets the same verdict wherever it comes from. The schedule is curated by relevance, not by sympathy. If that makes us awkward for every side, it's a good sign.
The four labels
The labels are four and public. There's no black box deciding for you: every verdict carries with it the elements to check it yourself.
Transparent correction
An automated system can be wrong: an imperfect transcription, a source missing from the index, an ambiguous number. We don't pretend to be infallible. When a verdict is wrong we correct it openly, and we improve the source index so it doesn't happen again — the path is documented on the Improvements page. Every verdict re-examined with a different label appears in the public corrections register.
Transparency of the method, and its limits
We explain how the system works, what we check and what we leave out (opinions, promises, rhetoric, well-known facts). There is, however, one stated limit: we don't describe the specific models and heuristics with which we extract and judge the claims. Not out of vanity, but so as not to help anyone who might want to game the system — the robustness of an automated fact-check also depends on not making it easy to fool.
One job, a thousand viewers
Each broadcast is verified only once: the fact-check timeline is computed once and streamed to every connected browser. The same verification serves a thousand viewers or one: we compute it once and show it to everyone in real time.
Privacy and data in Europe
Audio is transcribed on our servers. For extraction and fact-checking, text may be processed by external language-model providers, as described in the Privacy Policy. Core infrastructure and account data are hosted on servers in Germany. This does not exclude processing by the external providers listed in the Privacy Policy. We use technical cookies and, with consent, analytics cookies; no advertising cookies. Revolut handles card data. Details in Privacy and Cookies.
Verdict signing
When a signature proof is available on the permalink, you can verify the signed content's integrity offline using Ed25519. The signature establishes integrity, not the accuracy of the verdict.
Official public key (Ed25519, hexadecimal)
1719e2638a2cc0ec8d552d71dca52197f934e0a66b17e6364f1eed7a1d428e01
Canonical form of the signed content
JSON with sorted keys (sort_keys), compact separators ("," and ":") and non-ASCII characters escaped (ensure_ascii); the signature is base64-encoded.
Offline verification (Python, cryptography library)
import base64, json
from cryptography.hazmat.primitives.asymmetric.ed25519 import Ed25519PublicKey
proof = ... # JSON da /v/{id}/proof
pub = Ed25519PublicKey.from_public_bytes(bytes.fromhex(proof["public_key_hex"]))
blob = json.dumps(proof["payload"], sort_keys=True, separators=(",", ":"), ensure_ascii=True).encode()
pub.verify(base64.b64decode(proof["signature_b64"]), blob) # nessuna eccezione = firma valida
The full versions: Ethical principles · The model · Technology. Corrections and milestones on the Improvements page.
Document generated on 07/10/2026 · https://cryterio.it/metodo