Ask someone who has been dead a very long time.
Put a question to Marcus Aurelius about duty, to Ada Lovelace about her Engine, to Sun Tzu about a rivalry. They answer in their own voice — and beside every answer sits a ledger grading each claim: documented, attributed, inferred, contested, or beyond their world. Because the thing a convincing voice is worst at is telling you which of those it is doing.
The recording ships with this page and replays through the real renderers — the voice, the attestation ledger and the checks — with no account and no credits. It includes a question from after the figure's lifetime and a request the app refuses, because those are the two cases worth seeing before you spend anything.
Choose who you are speaking with
Each card says how that figure's record reaches us, because it changes what any of them can honestly claim. A private notebook, a student's memoir, an interpreter's transcript and an opponent's chronicle are four different kinds of evidence, and this app grades against the one that applies.
Meeting a figure, the rule checks, the attestation audit and the recorded conversation are free and run in this browser. Only a question costs credits, and each answer shows what it cost.
Your conversations
Why the grades are relative to the figure
The tempting design is two grades: they said it, or we made it up. That taxonomy is not neutral. It is calibrated on people who wrote books, kept letters, and had their papers preserved by an institution that thought them worth preserving — which is to say disproportionately literate, male, European and recent. Under it Marcus Aurelius scores well because a private notebook survived, and Sitting Bull scores badly because his words reach us through interpreters, reporters and the army he was fighting. That is not a difference in how well we know what they thought. It is a difference in who owned a printing press.
So the grades here are relative to a declared property of each figure: how their record survives at all. For someone who wrote nothing, the strongest honest grade for a first-person claim is attributed, and the ledger names the intermediary. That is not a demotion — it is the accurate description of the evidence, and applying it evenly is what stops the well-archived figures from quietly reading as more credible than the well-remembered ones.
What the audit cannot do
The check in your browser can confirm that a grade is one of the five, that it is consistent with how this figure's record survives, that a question from after their lifetime was marked as such, and that no long verbatim block was smuggled through. It cannot verify that a claim is true, and it cannot reliably count how many distinct assertions a paragraph contains. So when it reports assertions as unaccounted for, that is a sentence-counting heuristic and it is the weaker of the two measures — a reply can carry one claim across four sentences and be perfectly well grounded. It says unaccounted, never wrong.