Resources

Verified · Cited · Scored

How the truth engine works, what a real citation has to prove, and why a contested number must never be averaged. Every piece ships with the output format it describes.

AI that cites sources

An AI that cites sources — and what that actually has to mean

Most 'cited' AI answers cite after the fact. Here's the difference between a footnote and a verifiable chain: opened sources, ranked tiers, and a score you can audit.

6 min read
AI that admits it doesn't know

An AI that says "I don't know" — and why that's the hard part

Hallucination isn't a knowledge problem, it's an incentive problem. How a hard-coded admission rule and a confidence floor stop the model from filling gaps.

5 min read
AI confidence score

The AI confidence score, explained properly

A 0–100 score on every answer, with the basis in one line. What each band means, how it's earned, and why an unresolved contradiction caps it at 70.

5 min read
AI contradiction detection

Contradiction detection: keeping both numbers instead of averaging them

When sources disagree on a death toll, a vote tally or a financial figure, averaging is a fabrication. How competing values are logged side by side.

6 min read
exportable AI reasoning trace

The exportable reasoning trace: every search, every page, as JSON

One click exports the full trace: exact queries run, raw tool outputs, the answer, and the verdict block with sources and contradictions. Free-trial answers included.

4 min read
offline AI knowledge mode

Offline knowledge mode and pinned context: two power-user switches

Freeze the model to training data and your own documents, with the network off. Pin material that outranks the model's general knowledge — and hear about it when they conflict.

4 min read
unbiased AI

"Unbiased AI" is the wrong ask. Auditable AI is the right one

No model is neutral. What you can actually demand: visible sources, ranked evidence, logged contradictions, and no ideological refusals or unsolicited moralizing.

5 min read
fact-checked AI

Fact-checked AI for researchers, journalists and analysts

A workflow for people who get fired for a wrong number: primary-source ranking, contradiction logs, exportable traces and an AI that refuses to fill gaps.

6 min read

Founding member · $9.99/mo billed yearly · first 100 members