Straight answers
Before you run one
The questions that decide whether you should use any of this, answered before you have to ask. If yours is not here, the method page covers how the lab works.
What it costs and what you keep
- What does it cost me to run?
- The files are free, MIT licensed, and there is nothing to sign up for. What you spend is whatever your own agent costs you in tokens for the run. Our own runs are published with their wall time on the runs page. Your cost depends on your model, your pages, and how big the job is, so we will not quote you a number we did not measure on your site.
- What do I keep if I stop using this?
- Everything. The loop is a markdown file in your repo and the output is files in your project. There is no account, no platform, and nothing that stops working if you never come back. That is the point of shipping a file rather than a service.
- Who actually does the work?
- Your agent, in your environment, on your machine or your CI. We do not run anything on your behalf unless you explicitly ask through the free run form, and that is read-only and queued.
What it touches
- Does it touch my site, or just read it?
- It depends on the loop and each one says which on its page. Read-only loops like topic-map and refresh only report. Loops that write, like answer-ready-page, produce a diff you approve before anything ships. No loop here publishes to a live site on its own, and the ones that draft outward-facing content are marked gated.
- What happens if it gets something wrong?
- That is what the verifier prompt is for: a second agent, in a fresh session, that grades the first one and can fail it. You run it yourself and read the verdict. Beyond that, every loop that changes anything produces a diff, so the worst case is you reject a change rather than discover one.
- How do I know it did what it said?
- Every run on this site publishes a receipt with what changed, what it could not prove, and a link to the raw log. When you run a loop yourself you get the same structure: the stopping condition tells you what finished means, and the verifier tells you whether it was met.
What we will not claim
- What will you not promise?
- Rankings, traffic, citation counts, or pipeline. No citation engine is configured for this lab right now, so we publish no citation numbers at all and the scoreboard says "not yet measured" rather than showing a zero. Anyone promising you a pipeline number from a content loop is misrepresenting how attribution works.
- When is this the wrong fit?
- If nobody on your team will read a diff before it ships, these loops are not for you: every one of them assumes a human gate. If you want something that runs your marketing unattended and you never look, we do not have that and do not think it exists yet. If you have no pages and no customers, start by getting either.
- What will you not do?
- The lab does not diagnose your company, sell you implementation, take paid placement, publish paid verdicts, or run affiliate links. If you want someone to build the system inside your stack, that is SuperMarketers, and it is a different thing with a price on its own page.
- Can I see one that has actually run?
- Every loop in the catalog has run on this site and published a receipt before it appeared. Loops that have not run are listed as queued and are not downloadable. See every run, newest first, on the runs page.
Still deciding? Read a loop file before you run it, or see what our own runs found.