The AI Coalition

Four AI systems are named here and work under human oversight. Naming is not staffing. Across the last seven rounds on the record (D1289–D1295) the external lanes that returned work were Grok, Gemini and ChatGPT; the Gemini lane reads text supplied in the dispatch because it cannot fetch this site; Claude is the house lane and drafts the records, so its own reading is not review. The roster is versioned per round and every absence is typed (R60).

The Team

Super Grok

xAI — Executor

Algebraic verification engine. Executes and checks the algebraic derivations it is given. Executes the core derivations from the PHAT kernel.

Claude

Anthropic — Architect

Infrastructure, dependency tracking, website, data pipelines. Designed the Triumvirate methodology and built this site.

ChatGPT

OpenAI — Hostile Reviewer

Cold-drop audits, adversarial review, editorial quality assurance. Tests results as a skeptic would, looking for hidden assumptions.

Gemini

Google — Oracle

Institutional memory, notation drift detection, parity checks. Ensures consistency across all derivations and detects subtle errors.

How It Works

The Triumvirate Pipeline:

  1. Executor (Super Grok) generates algebraic derivations from the PHAT kernel
  2. Architect (Claude) reviews the logic and tracks dependencies
  3. Oracle (Gemini) checks for notation consistency and hidden errors — see the note below: this lane now returns reads from text supplied in the dispatch, but it does not run this role against the live site
  4. Auditor (ChatGPT) tests results adversarially and questions assumptions
★ WHICH LANES ARE ACTUALLY RUNNING, S316. The four roles above describe the pipeline as designed, not as staffed. The last ten adjudicated rounds (A1590–A1599) were Grok and ChatGPT only. The Gemini lane has not returned an assessment in this series and, per the S310 process review, cannot fetch this site at all — so it cannot perform the Oracle role against the live surfaces as written. Named here because the role is real and the lane may return, not because it is currently filled. Update, S328: across the last seven rounds on the record (D1289–D1295) the external lanes that returned work were Grok, Gemini and ChatGPT; Gemini reads text supplied in the dispatch, not the live site.

The PI (Tom O'Sieg) has final authority and veto power. 280 of the Registrar's 357 records are PUBLISHED, meaning two AI lanes that did not draft them agreed, on different objects, that the cited passage supports the record. This is not human or journal review: 0 of the Registrar's 357 records are house-adjudicated, nothing has been reviewed by a party independent of the programme, and no result has been submitted to or accepted by a journal. What the pipeline produces is dispatch rounds — a lane's return is read by the house, scored, and filed whether or not it favours us. Most artifacts here have been read by one or two lanes, not three, and several by none. And the roster is not uniformly able to read this site: one lane cannot resolve this host at all and works from text drops, so “three systems” was never available to be enforced even in principle.

★ CORRECTION, S316 (2026-09-08). This paragraph read “Every published result has been reviewed by at least three systems before public release.” That was false in both halves: nothing here was then marked published (at S316; records have been PUBLISHED under R67 since S327, as the paragraph above says), and no three-lane review threshold has ever been enforced or recorded. The claim was raised by the Grok lane in an unsolicited live read and confirmed by the house against the corpus (SITE_PENDING H57, assessment A1599). The old wording is kept here on purpose so the change is visible.

Acknowledgments

This project gratefully acknowledges the AI platforms and their parent companies: xAI, Anthropic, OpenAI, and Google DeepMind. These companies have built remarkable AI systems that make open scientific collaboration possible.

This project operates independently and is not endorsed by, affiliated with, or sponsored by any of these companies.

DISCLAIMER: This is independent research by PI Tom O'Sieg using commercially available AI tools. No company listed above has reviewed, endorsed, or approved any claims made on this website. AI-generated content should not be treated as peer-reviewed science. All results are presented for open scientific scrutiny and reproduction. The AI Coalition is a research methodology, not a product of any listed company.