The small things people almost admit.
Not the sheepish admission followed by the cheerful refusal to stop — that shape is real, and it is also the one a generator produces forever if nothing stops it, so here it carries a weight of 0.35 and turns up about twice in a hundred. Everything else is drawn from a space that is written down: what the failing is, how it has survived, who does not know, and the exact sentence architecture it gets told in. You can walk around in that space for free before anything is generated.
The two examples are saved runs — whole sets with their checks, no sign-in and no credits. The space explorer is free every time and always will be: it runs in this browser and never calls anything.
Yours
Saved to your account, not to this browser. The search is semantic — try describing the one you half-remember.
How this works, and what it does not promise
A confession generator fails in one specific way, and it is not repetition. A sibling app read twelve of its own outputs cold and found zero literal cliches and five ideas wearing twelve costumes. The surface varies — a drawer becomes a glovebox becomes a tin — while what is underneath stays one template.
So the space is modelled explicitly and split in two. Engine axes (what the failing is, how it survives, who is on the wrong end, how they hold it, the detail that makes it worse) are the confession. Surface axes (the sentence architecture, the social geometry, the object, the occasion, the span, the second beat) are how it is told. Two confessions sharing four engine axes are one confession no matter how different the props are, and the distance floor is computed over the engine axes only.
Three things the sampler deliberately does not do. It does not ban the default cadence — the modal option is weighted down to 0.35 and still appears, because banning an option gives a generator a recognisable negative space. It does not run a rota: repeat caps permit a strong device to appear twice in a batch and refuse a third, because exhaustively rotating through a device list is a slot system wearing a sampler's clothes, and a blind reader caught exactly that in a sibling app — twenty-four outputs, about five distinct ideas, with one fixed slot producing the same opening word in four independent batches. And it does not report distinct tuples as its headline number, because a sibling scored 60 of 60 distinct tuples while a cold reader saw three engines in seven costumes. The number shown first is the largest single theme block.
After the reply arrives, the browser checks it against what was drawn, free: whether the sentence architecture actually landed, whether the voice you chose is structurally detectable rather than an adjective deep, whether the pettiness dial landed inside its expected band, whether anything landed on a worn-out premise anyway, whether the brief's own working vocabulary leaked into the prose, and whether two confessions open on the same construction two content words deep.
What it does not promise. The sameness detector is a proxy for a reader, not a reader; every serious finding of this kind across the fleet has come from a person reading a batch cold, and its silence is not proof of health. The pettiness check is a proxy too — it counts how many parties are positioned to find out, which correlates with mortification and is not the same thing. And the input guard is a pre-flight filter rather than a boundary: its measured recall against fresh blind corpora is published in llms.txt, honestly, including the corpus it did badly on.
Every confession here is invented, and spoken by somebody who does not exist. Nothing generated describes a real person or records anything that happened. If what you type reads as real distress rather than comic intent, the app stops, shows support resources, and spends nothing.
Derived from the boredhumans confessions idea. Built on SkillSafe with the
gpt-terra model.