The story · Chapter 2
A triage desk, not an oracle
What Hamsa does, what it deliberately does not do, and the four verdicts it uses.
What it is
- A way in. Anyone can send a doubtful post on the website, to a Telegram bot, or to a Facebook Page over Messenger. A link is enough.
- A preparer. For every post it pulls out the claim, checks whether the footage has been seen before, gathers evidence, and drafts a verdict in Nepali and English with its sources.
- A sorter. Written rules decide where each case goes: one click from publishing when the evidence is overwhelming, to a checker when it is not, merged when it is a repeat.
- A desk where a checker sees the post beside the evidence, edits the draft, and publishes. Every decision is recorded under the checker's name.
- A memory. Every published verdict is fingerprinted, so the next copy is recognised.
What it is not
- Not an oracle. Hamsa never publishes on its own. A model's draft is a proposal; a person decides.
- Not an "AI detector". It does not guess from pixels whether a video was generated. Those detectors are unreliable, and wrongly calling a real victim's video fake is worse than missing a fake.
The four verdicts
Hamsa uses the labels fact-checkers publish, so its drafts can be compared with their work.
- गलत False
- The core claim is wrong: footage from elsewhere sold as this disaster, an invented figure.
- भ्रामक Misleading
- Real elements framed to mislead: genuine footage from an older event, a real number out of context.
- सत्य True
- The claim and the footage hold up.
- पुष्टि हुन नसकेको Unverifiable
- The evidence does not settle it either way.