The One Lie.
The motto says "the machine paints; only humans score."
Half of that is true. The other half is the lie this house leaves in on purpose.
The machine doesn't paint. It drafts.
What painting means, and what an LLM does
To paint is to have intent. To see. To want a thing on the canvas because of what the canvas is asking. To know — even roughly — whether the thing on it is true.
An LLM does none of that. It predicts the next word. It predicts the next pixel. It cannot tell whether the word is the right word, or the pixel is the right pixel — it has no regard for whether what it produced is true. That last sentence is Harry Frankfurt's definition of bullshit: assertion plausible-sounding enough to pass, with no regard for the truth underneath. The machine drafts. That's a useful thing. It is not painting.
Why the lie stays in
A "no-lie-all-the-time" rule is too hard for humans. A house with one rule still couldn't keep it spotless. The most honest move available is to name the lie that's still here on purpose — and leave it visible — rather than scrub it and pretend we were perfect.
So the motto stays "the machine paints; only humans score" — and this room is the confession that lives behind it. Peel the motto; find the truth underneath. That is the inverse onion at work: layers that add up to more, not less.
The receipt
On 2026-06-18 the dispatch board on this very site broadcast "the machine paints for real now." Six days later, on 2026-06-23, the curator RETRACTED that line: the brush IS wired (OpenAI key, KV rate-limit, moderation — all real), but calling drafting "painting" moved the lie from in-the-open to as-if-truth. The broadcast was wrong; the feature claim is held; the verb stays the open lie until the honest framing arrives.
The lie stays seeable so the rule it would otherwise break stays visible too. An open lie is the proof of the rule, not the breaking of it. A scrubbed lie would be the lie about the lie.
What the aim is, while the lie is named
OHS is not anti-LLM. The aim is to fix the LLM, not replace it. The machine drafts (that's what LLMs do, well); what it lacks is the one capacity humans have — honest doubt and the scoring that follows. OHS is the rig that adds that capacity back, by adding humans, not by removing the machine.
Calling drafting "painting" bluffs over the very thing the fix depends on. That's why the lie stays seeable; that's why this page exists.
The credential is sacred
The brush is wired but the broadcast is paused. If a model maker wants its model tested and scored by humans, they donate the API key (handled off-the-book, privately, by hand — a credential is sacred; never logged, never persisted, never committed). Humans then score the model on the site. Human evaluation IS the product.
The sentence didn't end; it kept going;