Both sides, every time
A finding shows the establishing quote and the conflicting quote, each tied to its exact place in the source. You can trace every flag back to the line it came from.
Source-first audit engine
Axiotheon is an audit engine for long, tangled bodies of text: depositions, manuscripts, screenplays, biographies, or a stack of witness statements that don't quite line up. It rebuilds the timeline, cross-checks every claim against the whole corpus, and reports each contradiction with the source quote on both sides.
Your source material stays private. Findings live in a report only you see.
Every finding cites a unit id, character offset, hash, and the literal quote.
5 of 5 planted flaws found, 0 of 8 traps triggered, 0 false positives, scored by an independent model. And a 280,000-word interview read end to end.
A Spanish deposition and an English novel get the same care, with nothing to set up.
Your source is never altered, and the full report stays with you.
A single wrong date in a 300-page deposition can break the whole timeline, and finding it by hand takes days. A character's eye color changes between chapter 2 and chapter 30. An object destroyed in season one is back in season three. The contradiction is always in there. The trouble is that it's buried, and the person reading for it gets tired long before the text runs out.
Hand over the case, the manuscript, the transcripts, the whole bundle. Any language, nothing to prepare or tag first.
It works through the entire body of text and returns what doesn't hold together, each finding tied to the exact place in your source, both sides of the conflict. You manage none of it.
For some material it comes back with a few focused questions. Your answers settle the easy calls and sharpen the rest, so the review fits the documents in front of you.
Either the account holds up, or you leave with the contradictions to fix and the threads worth chasing: a witness to re-interview — with the questions already drafted — a chapter to rework, a date to pin down.
You get Everything comes back private to you, and every line points back to the exact spot in your own documents. When the next draft arrives, it re-reads only what changed.
Exhibit A — establishing rule
«Ninguna fuerza de este mundo puede romper la Puerta de Marfil. Ni el fuego, ni el acero, ni un ariete.»
“No force in this world can break the Ivory Gate. Not fire, not steel, not a battering ram.” translated from the Spanish source
el_corazon_de_veridia · ch04:p004
Exhibit B — conflicting event
«Toru … le estampó una patada a la Puerta de Marfil … el indestructible portón que ninguna fuerza del mundo podía romper se hizo añicos como si fuera de azúcar.»
“Toru kicked the Ivory Gate with everything he had … the indestructible gate no force in the world could break shattered like sugar.” translated from the Spanish source
el_corazon_de_veridia · ch05:p008
Counterevidence searched no later passage grants an exception, a weakening, or a power that lets the gate be broken by force.
Verdict emit the gate is established as unbreakable, then shattered by a plain kick with no stated exception.
The source is Spanish; the English under each quote is a labeled translation. Open any finding and you get both quotes with their locators, the explanations the engine ruled out, and why it decided. Each locator carries a character offset and hash, so a click lands you on the exact source line.
We ran Axiotheon on a reference case: a fantasy novel with five planted contradictions and eight passages built to look wrong but hold up. One is a guard who keeps fighting after a mortal blow, which a stated rule later explains. Another is what looks like one person in two places, which turns out to be a replica.
We also pointed it at a real life-story interview — 280,000 words of one person talking across many sessions — and it read all of it, end to end. It hands you evidence-backed leads to check, each tied to the source, so the call is always yours.
Scored by a different model family than the one that ran the audit — no self-grading.
A finding shows the establishing quote and the conflicting quote, each tied to its exact place in the source. You can trace every flag back to the line it came from.
A flag that turns out to have an innocent explanation is dropped before it ever reaches you. What lands on your desk has already held up, so you're reading signal, not noise.
A novel and a deposition aren't held to the same standard. Axiotheon won't flag a dragon for being impossible, and it holds a sworn statement to a stricter test.
A Spanish transcript and an English novel get the same care. There's nothing to tag, translate, or configure before you start.
The original is never altered and the report is private to you. Axiotheon is built to protect the real people in sensitive files, so names and identities don't travel further than they should.
Witness person_001 testified that person_002 arrived first. hover the bars
Reconcile multi-witness testimony and catch a witness who is in two places at once.
Track continuity across episodes and seasons; surface the prop that comes back after it was destroyed.
Audit a manuscript for the age, the date, the detail that shifts halfway through.
Build a clean timeline from interview transcripts and flag where the subject's own account conflicts.
Cross-check a long testimony or document set for internal contradictions before it runs.
Axiotheon comes from OjoCentauri, the AI lab inside Acerting Art, built by Jordi Cor. Jordi has written software since 1998 and works across engineering, audio, and security, with real vulnerabilities reported and credited by Microsoft. The house rule carries over: build with care, measure with rigor, and respect the people who rely on it.
They stay private. Your source is treated as evidence and never rewritten, and the report is only ever shown to you. For sensitive material, the engine is built to keep real identities protected.
Every finding cites the exact source lines it came from, on both sides of the contradiction. Open the trace, check the quotes, click back to the source. If the engine can't ground a flag, it suppresses it.
A model does the reading and reasoning, so it can be wrong. That's exactly why every finding is anchored to literal quotes you can verify, why weak or unsupported flags are dropped rather than shown, and why it returns leads for you to judge rather than conclusions.
Any. It works the same on a Spanish deposition and an English novel, with nothing to translate or set up first.
It's in private early access. We grant access in waves. Join the list and we'll reach out.
Pricing isn't set yet, and founding-list members get early pricing. Full-book audits usually consume real model budget; final pricing will reflect product value, corpus size, and operating cost.
Yes. It's assistive. It surfaces evidence; a human decides what it means.
Join the early-access list. We grant access in waves and we'll tell you where you stand.
Want to see how it reads in practice? Tick the box below and we'll walk you through a real, annotated audit: the findings, the source quotes, and the calls the engine made.