Frequently asked questions
Is this an AI fact-checker?
No — it's a verification workspace. Claim extraction is deterministic heuristics; verdicts derive from human-graded evidence. Optional AI assistance can summarise, but judgement and citations always trace to real, dated sources a human linked.
Can it invent a citation?
There is no code path for it. Sources enter the system two ways: the logged search adapter (with retrieval dates) or manual entry by a user. A claim with no sources reads 'unresolved' — the report says so plainly.
What search engine does it use?
A pluggable adapter. Out of the box (no key) it uses a clearly-labelled local demonstration corpus so the whole workflow can be evaluated offline; add a Brave Search API key for live web results. Other providers are a one-function swap.
What does 'mixed' mean?
At least one linked source supports the claim and at least one contradicts it. That disagreement is the finding — the report shows both sides with dates.
How are search costs controlled?
Per-workspace hourly rate limits, every query logged with its result count, and manual sourcing always available as the zero-cost path.
Does it store full copies of articles?
No. It stores metadata, your notes, and a short excerpt (400 characters max) — enough to locate the passage, short enough to respect the source. The link is the citation.
Can clients see each other's work on the Agency plan?
Never — client workspaces are strictly separated, enforced server-side with cross-tenant tests in the release pipeline.
What happens after my trial?
Projects stay intact and exportable. Pick a plan to keep working; nothing is deleted for 90 days.