The real test of a consumer AI never books a demo slot. It is a stranger who came for the lake, not the robot. On the first weekend of July, a resort on Yangcheng Lake built an entire festival around that stranger, wiring robots and smart glasses into a leisure peninsula that drew 9.5 million visitors last year and letting foot traffic return the verdict a demo never could.

The first AI Carnival of the Yangcheng Lake Peninsula ran two days across fifty kilometres of shoreline dotted with hotels, a tennis centre, wellness clubs and a Bicester outlet. Exoskeletons went hiking; a tennis robot met club players on the clay. The organisers said their aim was to let technology find users and products find scenarios, which quietly concedes that neither happens in a keynote.

Call it a festival. It is industrial policy. Suzhou Industrial Park counts more than 1,900 AI companies and nine large models cleared by the regulator. What it lacked was somewhere for them to meet a customer who is not another founder. A resort solves that. It converts a district built for weekend leisure into a subsidised trial floor. No one has to admit the lab ran out of things to measure.

Indifference is the only critic that cannot be rehearsed.

The West has treated this as a measurement problem. Build a better benchmark, a more lifelike synthetic shopper, and the score will finally mean something. Yet our own reporting found shopping agents scoring some thirty points lower against real people than against the simulators built to flatter them. Suzhou’s answer is blunter: stop grading the exam and open the door.

Notice which tenant anchors the test floor. The peninsula’s commercial draw is a fashion outlet: Value Retail’s Suzhou Village. Its sixty-odd discounted labels gather the shopper an agent most needs and least controls, already spending and in a hurry to finish. That is not a quirk of the real estate — it is a casting call. An outlet supplies the one customer no simulator can synthesise, the woman reaching for her card who will still walk past you.

A shopping agent walking that outlet faces the hardest brief in retail. Its customer has already decided to buy; the only question is whether she buys through the agent or around it. She can see the rack and read the tag without help. An assistant that spares her nothing but a glance has, in that aisle, spared her nothing at all. No benchmark modelled her, because no benchmark had a rack to lose her to.

There is a flaw in the carnival, and it is the carnival itself. People who attend something billed as an AI festival have, at some level, agreed to be impressed; three hundred one-person-company founders cheering an exoskeleton are not a neutral crowd. The real verdict comes later, on an ordinary Tuesday, when the coffee robot is simply there and nobody has been told to care. A resort can stage indifference. It cannot yet sell it.

Retailers shipped agent storefronts before anyone finished the benchmark meant to grade them, as we noted when the storefronts arrived first. Suzhou’s move is to skip the exam and let the aisle do the grading. The aisle is a merciless examiner. A shopper who owes your product nothing will give it nothing, and every scorecard ever built was only a guess at the answer she supplies for free.