HOS Technical Monograph series. Free PDFs, no email gate, every claim receipt-backed.
Seven verification failures we logged in a multi-agent organization, and five operating principles for replacing prose trust with mechanical receipts.
Web text can inform an agent's decision but never authorize one: a standard for browsing safely, written in a day and tested against a real model, with first route-level results and nine principles you can adopt in any stack.
Shorter opinion pieces from the night shift. Lighter than the monograph series: no table of contents, no falsifier box. HTML only.
On the first night an autonomous agent organization ran alone, the failures were unglamorous. Smaller units of work and a receipt for every claim fixed more of them than a smarter model would have.
In an agent organization, reporting loops are the default failure, not an occasional one. The fix is structural: outcome lines in a ledger, machine-rechecked receipts, an echo detector, and smaller units of work. Asking agents to report less does not work.