Skip to content
Hugin
← JournalJournal

How Hugin reads public evidence

methodtransparency

Hugin turns a public post into a structured, linkable report. This is a short walkthrough of what that actually means — and, just as importantly, what it does not do.

What a scan reads

Every scan works only from data the platform already publishes to the world:

  • The post itself — title, body, score, timestamps.
  • The visible commenter context at scan time.
  • Account metadata the platform exposes — age, public karma, suspension status.
  • Outbound links and any identifiers in the text (domains, wallets, Telegram/Discord handles).

There is no login, no private-group access, and nothing scraped that the platform doesn't already show a logged-out visitor.

How a verdict is made

Two layers, in order:

  1. Heuristics. Deterministic checks fire on structural fingerprints — zero-day account swarms, copypasta clusters, off-platform funnel language, reused identifiers across reports. These run the same way every time.
  2. A written verdict. The signals are handed to a language model that names the playbook and writes a short, human-readable summary. If that step is unavailable, the report falls back to a heuristic-only verdict — it never blocks the report.

A polished, sincere-sounding post does not override hard structural evidence. A well-written post sitting on top of a sock-puppet comment section is still coordinated activity.

Why publish reports at all

Public review is stronger when the evidence is organized. Instead of "trust me, this looks fake," a Hugin report is a link anyone can open — showing the accounts, ages, patterns, and language signals that informed the call. You can disagree with the label; the underlying data is right there.

Corrections

The platform data is what it is, but the playbook label and verdict wording can be wrong. Those are revisable — that's what the dispute/correction path is for.