Trust & transparency
How review works
This is the public disclosure for the review pipeline. The author walkthrough shows the product flow. This page states what actually reads a submission, what is stored, and what becomes public.
Three specialists, then a coordinator
A review is not one model writing a grade. Three specialist agents read the same submission independently. A coordinator then reads the work plus those reports and writes the unified assessment.
Math / Logic
Derivations, dimensional consistency, notation, and proof steps.
Scores: Mathematical Validity, Internal Consistency
Sources / Evidence
Citation verification, evidence quality, and linked supporting papers.
Scores: Completeness, Evidence Strength
Science / Novelty
Testability, novelty, and whether the claims can be checked.
Scores: Falsifiability (or the converted domain criterion), Novelty, Clarity
The coordinator does not replace the specialists. It resolves disagreement, states the publication recommendation, and lists what would raise the score. For a plain-language explanation of the five-point scale, see how scoring works.
Which providers read the work
Each specialist slot can run across more than one lab so disagreement is visible. The current lineup is configured in admin and typically includes Anthropic, OpenAI, Google, xAI, and Together / DeepSeek. Provider API terms govern that hop: submissions are sent for inference, not granted as training data by TOE-Share.
Cross-model disagreement is treated as a signal. If the specialists conflict, the coordinator has to say so instead of averaging it away.
Stored, published, and private
- Draft — only the author (and admins when debugging) can see it. It is not in the sitemap and is not a public paper or framework URL.
- Reviewed, unpublished — the author keeps the report and timestamps. The work is not listed on the public papers or frameworks pages.
- Published — the public URL, sitemap entry, and (unless the author hides them) scores and specialist reports become visible. The review record stays even if the author later retracts the listing.
Author visibility toggles can hide scores, specialist reports, review history, or equations on the public page. They do not erase the stored review. See withdrawal for what can be removed.
What review is not
AI review is informational. It is not journal peer review, not a claim that a theory is true, and not an institutional endorsement. The panel scores structure, math, evidence, and testability inside the author's stated assumptions. It does not grade alignment with mainstream consensus.