Lab 03 jev · canvas · simulation

Jev moves a crowd

Building

A plaza with 1,000 people, each with a one-line biography. Type what just happened and a decision model decides, one call per person, who runs, who films and who steps in. No rules were written.

Click the plaza to set where it happens. The closest people answer first.
Click a person to read their line and their answer.
Same event, new spot: cached per person, so the rerun is free and instant.

waiting

reached: 0
calls: 0
cached: 0
time: –
cost: $0.0000

click where it happens

See one of the thousand calls, exactly as the server sends it

POST /api/labs one person jev · × 1,000 people

{
  "state": "Liam, 24, poet, with a bandmate, about to propose, anxious, wants to be famous, is a hopeless romantic",
  "questions": {
    "reaction": {
      "type": "choice",
      "instructions": "Something just happened in the plaza: \"the lights go out\". What does this person do next?",
      "criteria": {
        "flee": "gets away from it as fast as possible",
        "approach": "moves closer to see what is going on",
        "film": "pulls out the phone and records it",
        "help": "steps in to help or intervene",
        "celebrate": "cheers, dances or happily joins in",
        "ignore": "keeps doing what they were doing"
      }
    }
  }
}

The state is one biography and reaction is the only question, so the answer is a label plus a probability for each of the six. Before the fan-out, one more call reads the event itself to pick the scene:

{
  "state": "the lights go out",
  "questions": {
    "effect": {
      "type": "choice",
      "instructions": "Pick what best describes the atmosphere this event creates in a crowded plaza",
      "criteria": {
        "dark": "the plaza goes dark or the lights go out",
        "rain": "rain, storm or bad weather",
        "danger": "violence, fire, police, a crash or a medical emergency",
        "party": "music, a celebration, free stuff or a celebrity",
        "calm": "nothing dramatic, a small everyday thing"
      }
    }
  }
}

Both go through the same jev.classify() helper from Lab 02, with concurrency 64, a cache keyed by person and event, and a timeout per call so one slow answer never stalls the square.