Platform updates

Big multi-paper frameworks no longer time out mid-review. Reviews run asynchronously and large packages get a condensed pass that keeps section leads and primary evidence — so the panel can finish instead of refusing at the size wall.

Credit cost now scales with content size, and when a review needs more than your monthly allotment you can buy just that run via one-time checkout — no forced upgrade for a single oversized submission.

Claude, Codex / ChatGPT, and Cursor can connect to TOEShare over MCP — create drafts, import from URL, submit reviews, and publish without leaving your AI conversation. Prefer browser OAuth; agent tokens only for clients that don't support it. See /connect-ai.

Papers and frameworks listings now offer in-place Start here doors and topic chips from tags already on the page — filter the current list instead of bouncing to search.

Published papers and frameworks show a collapsed References section when verification rows exist — status counts in the badge, full list on expand. Owner Check/Re-verify stays on the dashboard.

Three public trust pages now explain the multi-agent pipeline, how authors can withdraw published work, and how Priority Defense challenges work — linked from Terms and the review bar.

Pure mathematics converts Falsifiability to Verifiability; first-principles derivations use a Necessity rubric. Physical theory, methods, and hybrid work keep the right criteria instead of getting dinged for the wrong kind of claim.

Citation checks now run automatically in review prep and feed a compact report to the sources specialists — so fabricated or broken identifiers get flagged before scores land.

A real paper with a dead DOI is no longer labeled 'fabricated.' Verification falls back to title/author search, and specialists are blocked from inventing arXiv IDs or copying uncorroborated math claims into the final assessment.

We published a multi-agent review of a viral proposed proof — honest highlights and lowlights, not a settled verdict — with a public review profile and homepage feature card.

Authors can connect a Zenodo account and publish validated papers or frameworks to an external repository with DOI-ready metadata, permanent-record warnings, and a visible publication history.

Reference checks now verify DOI, arXiv, CrossRef, OpenAlex, Semantic Scholar, official web docs, and LaTeX bibliographies. Each result shows a source, confidence level, and 'How verified?' audit trail.

Published papers can be reviewed again after improvements, keeping public scores and review context aligned with the current version instead of freezing feedback at first publication.

Tier discrimination, repeatability, prompt version delta, and bias isolation — all done. Scores are meaningful, repeatable, and not biased by model family. Full data is public.

Our system scored a Royal Society / MIT paper 2/5, flagging a missing quantum potential term. A physicist independently published the same finding on arXiv days earlier. Strongest validation to date.

Reviews now include equation-level warnings when math specialists flag steps that are stated without derivation or could invalidate downstream results. Visible amber badges on your review profile.

Interact with TOEShare programmatically — create drafts, submit for review, retrieve feedback, and publish from Cursor, Claude Desktop, or any MCP-compatible tool.

DeepSeek V3.1, R1, and V4 Pro models now in the specialist panel via Together AI. US-hosted, SOC 2 compliant, no data retention. More diversity = stronger consensus.

Download a 3-page branded review certificate with dimensional scores, specialist reports, and evidence. Send it to arXiv endorsers, journal editors, or collaborators.

Select 2–5 published frameworks and chat with AI across all of them at once. Compare approaches, contrast predictions, or synthesize across entire bodies of work — right from the frameworks listing page.

Reviews now include a mandatory red-flag checklist and code-enforced score caps. Inflated ratings are caught automatically, ensuring calibrated, defensible scores across all 7 dimensions.

The platform now monitors review panel health in real time. If a model fails mid-review, admins receive email alerts and affected authors get a free rerun automatically.

As of April 25, 2026 our review panel runs the latest flagship models from OpenAI, Anthropic, Google, and xAI. Four providers, always up to date, stronger multi-agent consensus.

Frameworks are now evaluated on their evidence roadmap - testable predictions, motivating observations, and decomposable claims - rather than penalized for lacking experimental data at submission.

Authors can now make published work private, control what visitors see on public pages (scores, reports, history), and manage visibility per-document from the dashboard.

AI reviewers now evaluate within your stated axioms. Declare assumptions, get reviewed on logical consistency.

Search across papers, frameworks, and authors with tabbed results, faceted filtering, auto-complete suggestions, and a global Ctrl+K quick search from anywhere.

Specialized AI agents (Math, Sources, Science) coordinated by a lead reviewer for deeper analysis.

Import papers directly from Zenodo or arXiv by URL with automatic metadata extraction.