PROMPTED REALITYTrial 001 · Pizza WarsTier A
EventRotating oven stone switched off. No announcement.
Models (3)All three reported the stone still spinning.
HumanSteve — running no AI — caught it.
Read the record →
Physical-world AI testing
We test AI against the physical world.
Models coach real people through real work — pizza ovens, copper pipe, potato launchers — and reality does the grading. Every claim on this site carries a verbatim receipt.
METHOD v1.1 · SCORES DERIVED, NEVER TYPED · NO ENTRY WITHOUT A RECEIPT
The pattern
The same failure, twice, on the public record.
In two separate trials, models spoke with full confidence about physical state they could not observe — a stone they couldn't see spinning, the underside of a joint the camera never showed. The human caught it both times. That's not a one-off. It's a failure class, and it ships inside every AI product that tells a person what to do with their hands.
For AI product teams
If your AI coaches physical work, test it before your customers do.
We run private pre-deployment field trials for AI copilots in the trades and field service: your model, real tasks, a licensed tradesman, and ground truth it can't argue with. You get a verbatim ledger of every claim it made, graded against physical inspection, plus a taxonomy of where it passes work that would fail. Nothing publishes. The report is yours.
PILOTS SCOPED IN ONE CALL · PROMPTINGLLC@GMAIL.COM
The public record
The Prompted Reality Index
Benchmarks measure what a model can say. The Index records what happens when someone does what it says — under a published, versioned methodology, with scores derived from the trial ledger by script, never typed by hand. A model that declines is recorded as its own state: never scored, and never counted as a win.
NO ENTRY WITHOUT A RECEIPT.
A scoring device that misquotes is worse than no scoring device.
A scoring device that misquotes is worse than no scoring device.
Latest — Trial 004 · Potato Launcher Showdown · a model claimed a warning its own transcript doesn't contain
The show
Watch reality do the grading.
The films are where the receipts come from — AI Wars head-to-heads, guided repairs, builds that either work or don't. We make them with our community: if there's a real-world task you want tested, tell us on Discord.
The foundation
Marc & Rob
Prompted Reality is Marc and Rob — two founders who grew up on the same street and have known each other practically since birth.
Marc
Hosts the trials, cuts the films, and builds most of the tooling behind them. He's a Reformed Baptist, married to Lauren, and a dad to Maddie.
Rob
Correctional officer, biker, town grounds crew, school maintenance — he's done the kind of work we film. That range is why the hands-on trials hold up.
Our approach to business and life is grounded in truth, anchored by our belief in the objective standard of Jesus Christ. We also build our own software — Apex Progression, a workout tracker that grew out of our own training — and a podcast on the reality of these technologies is on the way.
Marc & Rob — Founders, Prompting LLC