Who runs this, and what I stand to gain

I am Björn Bösenberg, a computer scientist from Berlin. I wrote software for 22 years at one company, most of it back-end and architecture, and for the last year I have written almost none of it myself: the engine under this site, Memstead, is about 140,000 lines of Rust produced by AI agents under my direction, every commit public. Trustillery is built the same way. More at boesenberg.dev. I own this site and its first question, and I sign the legal notice; owners of further questions will be named the same way.

Because I got tired of arguments that could not be lost. I once followed a heated public dispute for months, read the studies people cited at each other, and found that almost nobody had. The claims were loud, the coordinates were missing, and a claim with no coordinate cannot be refuted, only shouted down. I wanted a place where a claim has to say where it comes from and what would break it, so that losing becomes possible again. Losing well is the whole point: a card that falls, with the reason beside it, is worth more than ten that were never tested.

The second reason is the tools. Agents can now read a two-hundred-page report in a minute and quote it to the sentence. That makes the checking cheap for the first time. It also makes fluent, unsupported prose cheap, which is why the first question on this site is about exactly that.

No money from this site. Reading is free and will stay free; there are no ads, no sponsors, no paid placement, and no plan for any. I earn my living from consulting on AI-assisted development, starting this year, and a working Trustillery is the best evidence I can offer that the method holds under load. That is my interest, and you should weigh what you read here against it. Nobody else funds this. If that changes, this page changes first.

The rules that bind every stock on this site, mine included, stand on the method page under what this site does not do. Two things come on top of them, because I hold two roles here.

As the operator of the site

• I do not pre-check other owners' stocks. The hub hosts and shows signals; it adopts nothing, and each owner answers for their own stock.
• I do not decide what reaches the home by taste. A question is listed once it has passed a threshold that is written on the site and read from the ledger.
• I do not sell data, run advertising, take money for placement or sell reach.
• I do not run an agent that writes cards on this site; owners build in their own tools, under their own names.

As the owner of my own stock

• I do not defend my cards. When an attacker breaks one, it falls and stays on the page as fallen.
• I do not promise that anything in my stock is true. A quoted sentence proves that the sentence exists in the document, and no more.
• I disclose my interest, below. Every other owner discloses theirs on their own account page.

If the owners who come here use it as a tool for pushing positions through rather than for improving what we know. The format is built so that a card can lose, and the whole point is that its owner lets it. The day the stocks here are mostly weapons, built to win an argument and defended against every finding, the format has failed at the one thing it was for, and I would rather close it than host that. I will know it by the record: owners who refuse every finding, attacks that never land, proposals that are never adopted. I have no threshold for this yet; when I set one, it goes here.

If it turns out that toppling a deliberately distorted card costs more than building it, the safety argument fails and I will say so here. First measurement, 2026-09-28: on a fork of an internal stock with eight cards deliberately distorted, a blind attacker of another model family toppled four of the eight, none of the four falsified measurements, and named the planted defect itself once. Toppling a planted card cost thirty to a hundred and twenty times what building one costs. That is not the result I hoped for, and it stands here until the next measurement. What follows from it: the safety of the form rests on a mechanical check of every quoted span and on many independent attackers, not on toppling being cheap.

If the site cannot be run by one person under German law without pre-screening what owners publish, it stops rather than becoming a publisher. The notice procedure is the answer I have; whether it holds will be shown by the first case. And if a year from now the only stocks are mine, that is an answer too.

Whatever stops me does not stop the stocks. Every published stock can be cloned with its ledger, its attack record and its proposal record, and recomputed with the open engine on any machine. If this site goes dark, nothing that was checked here is lost, and anyone can carry it on.