The factory learns to see itself
Forty-five engines brought under one registry, one ledger, one map. Round-trip integrity proven across the whole population — every record readable, writable, and reversible by its own gates.
The root of the story: an honest build log of one non-coder governing an AI workforce — published as it happens, receipts attached. This page is the storefront window; the publication itself lives on Substack.
A few load-bearing moments from the ledger — the same record the workforce itself runs on.
Forty-five engines brought under one registry, one ledger, one map. Round-trip integrity proven across the whole population — every record readable, writable, and reversible by its own gates.
During live verification, a write that was actually correct got reverted by its own safety gate — byte-for-byte — proving reversibility in production. Trust is structural in both directions: the gate that can undo a mistake can undo a success, and that's exactly the point.
Offer priced against real market data. Stripe live. Notes on a Build published. The workforce's first outreach circulated by the owner's own hand — and its one dishonest draft was rejected by human judgment before it ever left.
One fabricated email address — caught by a bounce — was metabolized into law within hours: the Evidence-First doctrine. No fact without evidence; models locate, retrieval creates, deterministic validation admits.
The pipeline that cannot mint a lie: every contact route on every draft is now evidence-backed or honestly absent. The founding defect — the very address that bounced — is caught by the gate it created, forever.