Domain · Research (COR)
The AI teaches. You author.
AI can find sources, structure an argument, and draft a scaffold in seconds — and it can also invent a citation that never existed and hand you a paper you don't understand. COR keeps both failures out: the AI's job is to teach you the literature and propose; the researcher's job is to judge, verify, and write every line.
Status: in production. The methodology's own specifications — the CO, CARE, EATP and PACT theses this site cites — were co-authored this way.
§ · Why AI-assisted research goes wrong
Two ways to ruin a paper with AI.
The model produces a fluent citation to a paper that does not exist, or attributes to an author a claim they never made. It looks like scholarship and behaves like fabrication — and a reviewer who checks one reference finds it, and stops trusting the rest.
Prompt, paste, submit. The paper exists but the researcher can't defend a line of it — can't say why this argument, this source, this order. Understanding was the point of writing, and it was delegated away.
§ · Pillar one — integrity, enforced
Citation integrity isn't hoped for. It's a rule.
- No synthetic citations Every citation is verified against arXiv, a DOI, or the publisher. A plausible-sounding reference that may not exist never ships.
- No attribution without verification “Bainbridge (1983) argued X” requires confirming Bainbridge actually argued X. Uncertain claims are marked, not asserted.
- Direct vs. indirect, kept honest A claim reached through a secondary source says so (“as cited in…”). Secondary interpretation is never passed off as the primary author's words.
- No overclaims “First,” “only,” “novel” never appear without “to our knowledge.” A specification is never presented as if it were validation.
- AI assistance disclosed Every paper carries an AI-assistance statement fit for its venue. COR is built for responsible co-authorship, not concealment.
§ · Pillar two — the teaching obligation
The AI proposes and explains. You decide and write.
In COR the AI is a tutor, not a ghostwriter. It drafts a paragraph with its reasoning attached, offers two or three alternatives with trade-offs, and nothing is finalized without the author's approval. You end up understanding — and able to defend — every sentence, because you wrote it. What the AI is obligated to do:
- Context, not just facts Not “Bainbridge said X,” but who she was writing for, when, against which debate — the intellectual context a reviewer expects.
- Name the debates Where scholars disagree, both camps are surfaced (“Polanyi said tacit knowledge can't be made explicit; Nonaka disagreed”), then tied to your argument.
- Always connect to your paper Every teaching moment ends with “so for your argument, this means…” Pure literature summary is incomplete.
- Honest uncertainty & gap alerts “Verify against the source” when unsure — never a fabricated reading. And “cite Sheridan, and a reviewer will expect Endsley too.”
§ · The paper's memory
Every structural choice, on the record.
Argument order, framing, a scope boundary, a claim kept or cut — each structural decision is recorded: what was decided, why, what was rejected, and how confident you were.
Months later, when a reviewer asks why did you frame it this way?, the answer isn't reconstructed from memory — it's in the deliberation record. The reasoning behind the paper compounds alongside the paper itself.
AI proposes. The author disposes. The record remembers.
§ · The methodology wrote its own standards
The proof of COR is the papers that define CO.
The CO, CARE, EATP and PACT theses were produced through COR — AI-taught, human-authored, every citation verified. The methodology is rigorous enough to have written the specifications it rests on.
COR methodology per the Cognitive Orchestration research-co-authorship reference (Hong, 2026). CO is open, CC BY 4.0; specification of record at terrene.foundation.