When AI makes confident writing cheap, defensible claims become the work.

A workspace where every load-bearing claim must trace to evidence, a source, or an open question — and the export shows when text outran its support. The system does not bless sentences because they sound good.

Want to feel it before you read the argument? Read the researcher's journey → — a narrative walkthrough of one researcher's day inside the finished platform.

The problem

Research has a painful version of the universal AI trust problem. Drafts arrive fast; citations and evidence lag behind. Reviewers, supervisors, and integrity offices care about one thing: can this claim survive scrutiny?

The discipline

THREE ALLOWED BACKINGS

your own evidence;
a real source or citation;
an explicitly marked open question.

The system is not saying the science is true. It is saying whether the manuscript has outrun the record.

What you do in the workspace

Sources, evidence, claims, writing, review, and export stay connected — not scattered across five tools — and shared across your lab, with per-person attribution on every row.

Import and organize sourcesPaste a DOI or import your Zotero library with your own key — and keep what you're building on where you can see it.

Attach evidence to claimsLoad-bearing sentences earn their backing or get flagged.

Write with the record visibleDraft in the same place the evidence lives.

Run rigor review before exportUnresolved claims travel in the export — visibly, not hidden.

A research assistant works inside the same space — it reads your draft, questions weak spots, and suggests sources, running on your lab's own model key. It never promotes a claim's status: raising a verdict from "inconclusive" stays a human act, on the record.

Reviewed access

Evidence Foundry is the first research-facing vertical on the Seitiate substrate. The early evaluation path is browser-based, reviewed, and intentionally limited to work that is safe to use in an early pilot.

Apply at evidencefoundry.com → Already approved? Open workspace →

Pilot scope

Synthetic, public, or explicitly approved non-sensitive data first. Real institutional datasets require governance, access control, and institutional expectations to be clear before they enter the workspace.

Built on Seitiate — exportable research records on a trust substrate. Company overview · not required to use the workspace.