An independent verification shop for AI-built code, run in public by one person and a team of AI agents under adversarial discipline.
Trenyx reads software that AI agents wrote and tries to prove it wrong. The attack plan is hashed and timestamped before a line of source is opened, the read is done blind, and whatever it finds goes to the maintainer privately first. The pre-registration and the full record publish once the maintainer has fixed it or cleared it. The published engagements are on the audits page.
Who I am: SK, solo developer and researcher. I publish under the handle blu400. I built the engine here with AI agents working under spec-first, test-first, adversarially reviewed discipline, and I ran it on my own systems before anyone else's.
The trading records are the standing demo. Two paper books run here every week, every decision checksummed and append-only, which means when something goes wrong, I can't quietly fix it. When the system halts on bad data, refuses to trade, or catches me wanting to break my own rules, that publishes too. The bugs are the marketing.
Work with me: I verify what your AI agents built, and I stress-test backtests. Fixed scope, flat quote agreed before I read a line. Details here.
If you follow along you'll get: the audits as they publish, the weekly scoreboard on Mondays, me vs. the monkeys, including the weeks I'm losing to a monkey, and a chapter of the build story when there is one to tell.