What it is, where it came from, and why it doesn't work like anything else.
Apodicta is a verification engine: a decision system built on one structural rule — the model proposes; the code decides. An AI model is used for perception only: reading, extracting, suggesting where to look. It is never allowed to author a conclusion. Every verdict is computed by executable logic over admitted, quoted evidence, which is why every Apodicta answer arrives with receipts — verbatim quotes from named sources, linked, so you never have to take anyone's word for anything, including Apodicta's.
Apodicta is not a search engine (it returns verdicts, not ranked pages), not a chatbot (it composes answers from computed results, not from plausibility), and not merely an "AI checker." It is verification — of claims, comments, posts, articles, citations, arguments, and yes, AI answers — against the public record and against the text's own logic.
Apodicta was built by Jereme G. Lytle, a practicing criminal-defense attorney and former prosecutor in Colorado, for his own caseload. Criminal practice is where sloppy reasoning costs people their liberty: a citation must exist, a witness's statement must be checked against every other statement they've made, a theory must survive the evidence that actually bears on it — not evidence that merely sits near it. General-purpose AI could not meet that standard, because language models generate what is plausible, and plausible is precisely what a courtroom exists to test.
So the engine was built the other way around: formal, executable rules — burden, corroboration, source-independence, timeline coherence, theory selection, bearing — with the model confined to perception and every extracted fact bound to its citation. The engine runs on the attorney's own hardware, against a closed, curated library of legal authority and case-specific fact sources, because in legal work the source universe must be exact and confidentiality is architecture, not policy. It runs real matters measured in tens of gigabytes, daily. Three U.S. provisional patent applications (filed July 2026) cover the architecture, its enforcement chain, and the Interpretation layer — the governed seam where perception is admitted into reasoning by rule, never by resemblance.
The engine was never law-specific. The formulas don't know what a courtroom is — they know what a claim is, what evidence is, what bears on what, and what follows. Law was simply the first domain hard enough to force the architecture into existence.
Apodicta.net is the public-facing side of that same architecture, pointed at everyone's problem instead of one profession's: paste anything — a chatbot's answer, a social-media comment, a headline, a claim somebody made at dinner — and the engine checks it against itself (contradictions, overreach, broken logic) and against the open public record: encyclopedic sources, the scientific literature (PubMed), published papers (Crossref), attributed quotes (Wikiquote), and the open web. Where the professional engine uses a deliberately closed source library for exactness, the public engine uses the open record for reach — same gates, same doctrine, different source universe, each chosen on purpose.
Every admission decision is content-blind code: a sentence becomes evidence only by membership in the claim's own terms, with every named entity present — never by ranking, popularity, or resemblance. Every receipt is quoted verbatim and linked. Agreement across independent sources is counted, never assumed; sources are themselves swept for self-contradiction; and when nothing can be verified, Apodicta says exactly that — a claim that cannot earn a verdict is reported as unconfirmed, never dressed as true or false. The rules never change; only the record can.
Verify something right now — no account, no key, nothing stored — or take the browser extension everywhere: select text on any page, right-click, "Apodicta this." Questions: jereme@apodicta.net.
Apodicta — patent-pending (U.S. 64/117,521 · 64/119,086 · 64/121,444).