Confidence is key.
Every fact Denua surfaces is graded, quote-checked and sourced — so you know what's solid before you rely on it.
Traffic-light grading
Every claim marked green, amber or red — never a single vague score.
Deterministic quote check
Each quote is string-matched to its cited page. No model marking its own homework.
A source on every line
Document, page and the exact quote behind each fact.
OCR-aware trust
A fact drawn from a shaky scan can't read as "sourced".
A real bar to clear
Auto-verify only at ≥0.92 confidence and a verbatim match.
Flag, don't guess
Anything uncertain is surfaced for review, never silently accepted.
Trust you can take to court.
Denua grades, checks and cites every fact it produces — the difference between an answer and evidence.
Deterministic faithfulness
The quote is really there — or it's flagged
Every cited quote is matched, character-for-character, against the page it claims to come from — pure text, no AI judging itself. If it isn't an exact match on the cited page, the claim is downgraded from green to amber automatically
Provenance
Sourced to the page, every time
Each claim carries its document, page number and verbatim excerpt — click a fact, land on the exact line in the bundle. A fact with no source in the bundle is labelled "recorded as instruction", never dressed up as evidence.
Calibrated
Only near-certainty clears itself
A claim auto-verifies only when calibrated confidence passes 0.92, the quote is a verbatim match, and nothing was normalised. Everything else waits for a human — and a fact from a low-confidence page can never present as clean.
Confidence, quantified
3
3
Trust tiers on every claim
92%
92%
The bar to auto-verify
1 click
1 click
From any fact to source
See what your evidence is really telling you.
Grade, check and cite every fact — in one workspace built for scrutiny.
