Run three phone participants in one public lobby

lobby-trivia-3player.yaml in the humanish repository is a worked example of the external-public shared-world route. Copy it into your project's humanish/studies/ to run it. Three phone-sized participants play the same multiplayer lobby on a public deployment at once, and one Observer shows all three. Point subject.appUrl and subject.publicTarget at a deployment you own or operate.

What the study doesCopy Anchor Link

• The public site is the shared plane. humanish clones nothing, starts no subject sandbox and seeds no data.
• The participant marked host: true creates the lobby first. humanish reads the /lobby/CODE path from the host's browser and puts the code into the other participants' missions. Those participants join through the app's own Join flow, because visiting /lobby/CODE directly does not make a visitor a member.
• The study passes when all three participants reach the same /lobby/CODE (lobbyConvergenceDigest, a digest only) and their sessions overlap in time.

Make the participants phone-sizedCopy Anchor Link

The participants use the mobile (414×896) and small-mobile (360×740) presets. Without mobile emulation, a hosted desktop renders both at 500 CSS px wide, because Chrome refuses a narrower window, so the two presets lay out the same and only the heights differ. The participants also get no touch input, and the device pixel ratio stays in the prompt and the metadata.

Add mobile emulation to the study so each participant gets its preset's real viewport:

execution:
  desktop:
    fidelity:
      mobileEmulation: true

With it, each Chrome or Chromium participant on a mobile preset browses at its preset width (414 or 360 CSS px), with the preset's device pixel ratio, touch events and a mobile user agent. The run bundle records what the page reported under desktopGeometry.fidelity. Emulation is still not a physical phone. Mobile emulation lists the settings and the read-back.

Run itCopy Anchor Link

A dry run is the default. It costs nothing and checks the study's plumbing and its evidence contract without sandboxes or model tokens:

npx humanish run lobby-trivia-3player

For a live run, set mode: live in the study, then run it with OPENAI_API_KEY and E2B_API_KEY available:

npx humanish watch lobby-trivia-3player --dotenv .env.local

To watch from a phone, serve the run's Observer through an authenticated edge:

npx humanish observe --all --expose --tunnel ngrok --oauth google --allow-email you@example.com

watch --expose streams live desktops only on the computer-use route, so a shared-world run is watched through observe --all.

What the evidence claimsCopy Anchor Link

The operator attests to owning or operating the deployment (subject.publicTarget: { owner, authorized: true }); humanish cannot verify that. The run records shared-world attribution, N participants on one plane, and humanish verify checks that it claims nothing stronger:

• provenance is external-public, because humanish seeded nothing;
• there is no synthetic attestation, because a public site is not synthetic;
• the operator controls the plane, and humanish only observed it;
• there is no authoritative shared-state proof;
• concurrency rests on overlapping sessions and the observed lobby convergence.

One run says nothing about adoption, scale or repeatability.

Check the app before a live runCopy Anchor Link

The lobby mechanics in this study were read from the app's source at the time it was written: the create and join calls that lead to /lobby/CODE, the six-character code alphabet ABCDEFGHJKLMNPQRSTUVWXYZ23456789, and the rule that a direct visit does not join. A new deploy can change them. Check the two places the study depends on: the lobby-code pattern (/\/lobby\/([A-Z2-9]{6})(?:$|[/?#])/, which tolerates a locale prefix, a query and a hash) and the wording of the join mission the other participants follow.