Mainstream AI hedges. Forums don't cite. The documents take months to read.
The public story is what a chemical company, a court, or an agency was willing to say. The documents show what they did, and what they spent to keep the rest quiet. Deep Clearance searches those documents and returns the passage with the source still on it.
S1 is the page that shows the act. A dollar figure in a filing is S1 for that figure. A press release is S4, evidence that they said it, until another page shows the act. A conclusion stays as strong as the weakest passage it needed.
When a subject has been branded misinformation, deplatformed, or litigated into silence, the first pass is the suppression itself. Who paid for it, when it started, and what form it took come before the documents.
Major claims carry the conditions that would downgrade them, and the strongest account on the other side sits with the thesis.
Language models are trained toward the managed consensus. Deep Clearance tracks where those hedges fall.
Financial capture from 1812 to Jekyll Island. Seven historians, seven methods, and the same suppression sequence. Patent secrecy orders, enforcement asymmetry, trillions in unsupported Pentagon adjustments, and the accounting standard written to keep them dark.
Over a million primary source passages, under one standard, with the source attached to the answer.
Pay with Lightning. Connect over MCP.
The cost of suppression just went up.
Every plan searches the same documents and the same interviews. A plan buys days and a daily cap on queries. The cards are a sample of a larger index. See the plans.
What companies, courts, and agencies wrote down, plus books.
Interviews and talks, with the source named.
A reply is a stored excerpt with the source attached.
GitHub or Nostr.
An upgrade keeps the same key.
Model Context Protocol (MCP). Claude uses OAuth. Others use a Bearer key.
The passage comes back with the source attached.
A stored passage and the source it came from. Strength and depth are labeled when the index has them. If the index has no passage, the tool says so.
About 1,476,000 passages from company files, court records, released government files, patents, interviews, and short excerpts from books. Classified documents are outside the index.
Days on the index and a set of queries each day. Trial is 7 days and 50 queries a day, for 15,000 sats. Researcher is 250 a day. Control is 500. The documents are the same on every plan. After the day's set is used, a Lightning top up pays for more. Replaying a snapshot is free.
Sign in with GitHub or Nostr, pay the Lightning invoice, and put the key in your client. Claude uses OAuth. Cursor and other clients send a Bearer key to the MCP endpoint.