PIJAlance Magazine News
Intelligence

Our architecture

How we decide, and what each verdict means.

A verdict is only as good as the rules behind it. This page sets out ours: what each verdict means, how sure we are, how sources are weighed, and the checks every answer must pass.

The nine verdicts

Most claims are not simply true or false, so our answers are not either. Each verdict below has one meaning, applied the same way every time.

True

The claim matches what reliable evidence shows.

Example: a statement that a named agency confirmed an arrest, when the agency did confirm it on record.

Mostly true

Accurate in substance, but a detail is wrong, missing or out of date.

Example: the event happened, but the figure quoted is slightly off.

Misleading

Built from real facts, but arranged or framed to leave a false impression.

Example: a true statistic presented without the context that changes its meaning.

False context

Genuine content (a photo, video or quote) presented with the wrong time, place or circumstances.

Example: a real video from another country shared as if it happened in Nigeria this week.

False

The evidence shows the claim is wrong.

Example: a portal said to be launched by a government body that the body itself says does not exist.

Unverifiable

There is not enough reliable evidence to decide either way, yet. This is not the same as false.

Example: a claim no trusted outlet or official source has reported on.

Speculative

A prediction or opinion presented as fact. It cannot be checked until events play out.

Example: a claim about what will happen at a future election.

Satire or parody

Comes from a satire or parody source and was not meant as fact, but is being shared as if it were.

Example: a joke headline from a parody account passed around as news.

AI generated

An image, video, audio clip or text produced by AI and passed off as real.

Example: a synthetic voice note imitating a public figure.

How sure we are: confidence

Every verdict carries a confidence level. It is set by the strength of the evidence itself: how many independent sources there are, how trusted they are, and whether any of them disagree. It is never set by how certain the AI sounds.

High

Several independent, trusted sources agree, or an official body has confirmed it on record, and nothing credible contradicts it.

Medium

Good evidence, but fewer sources, or none from the very top tier of trust.

Low

Thin or conflicting evidence. Low confidence is only allowed with "unverifiable", "speculative" or "satire", never with true or false.

How sources are weighed

The safety checks behind every answer

  1. Every fact needs a sourceAny factual statement in the reasoning that does not point to a specific source is rejected automatically.
  2. Same incident, same place, same timeThe sources behind a verdict are compared by place and date in code. A story from another state, another country or months apart is removed, so a similar old case can never be mixed into a new claim.
  3. Background pages stay in the backgroundTopic pages that list many different stories, and fragments too short to show where or when, are never used as evidence. The full article is opened instead.
  4. A second, independent AI checkA different AI model from a different company reviews every answer without seeing the first verdict. Disagreement sends the claim to an editor.
  5. Failures never reach you as answersIf any step fails, the claim is held for an editor and marked "not yet confirmed". It never replaces an earlier, correct answer.
  6. Clear disclosureEvery answer says whether it was approved by AI or by a human editor, and links to all of its sources.

Seen something you're not sure about?Check it before you share it. It takes less than a minute.

Check a claim
Your Pocket NewsroomCheck a claim now