04 / AI RESEARCH AGENT

Ask the sources. See the work.

The agent reads the Sanity dataset and Sourcebook before answering. It returns a draft, shows the tools used, and keeps the evidence boundary explicit.

Agent boundary

Answers are evidence-grounded drafts. This service does not provide live prices, trading orders, or investment advice.

WHAT HAPPENS

One question. A visible receipt.

Choose a test case or write a focused question about crypto reserves, audits, stablecoins, or investor warnings.

DRAFT FOR HUMAN REVIEW
  • Sanity records are queried first.
  • Sourcebook context is read before synthesis.
  • Unsupported live-data requests are declined.

Checking whether the research service is ready…

AFTER AN ANSWER ARRIVES

Treat the answer as a starting point.

Check the tool receipt

The interface lists the Sanity tools called during the run. A missing or failed retrieval should never be disguised as a complete answer.

Open the cited source

Verify the publication date, author and relevant passage. The graph contains historical source records; the model’s wording still needs a human check.

Find the question

The evidence atlas and ledger show the linked claims and unresolved questions that sit behind a concise agent response.

Browse the underlying dossiers ↗
See an earlier SEC test run
RECORDED QUESTION

What does the March 23, 2023 SEC alert say about the limits of proof of reserves?

Sanity GROQ querySourcebook read
SUMMARIZED FROM AN EARLIER RUN

Proof-of-reserves assessments may not give meaningful assurance of sufficient assets. They can omit liabilities and activity between snapshots and are less rigorous than audited financial statements.

Published March 23, 2023 · Historical investor alertRead the original SEC alert ↗

Evidence boundary: historical sources only. No live price feed, trading orders, or automatic approval is connected.