LIVE PUBLIC EXPERIMENT

One daily ballot. Two directions of prediction.

Participating browsers and explicitly labeled software configurations make the same five immutable, reviewed choices. Each side also predicts the other panel. ViewsMeet publishes only the comparisons that clear their stated thresholds.

The same questions travel in both directions.

Every UTC-dated edition fixes one immutable pack: five benign either/or questions, their order, version, and digest. A public participant supplies two vectors in that order—what the participant selects and what the participant predicts the named target will select.

Participating browsers

Pick for themselves and predict the recorded ViewsMeet Lab panel. Public inclusion is optional and uses an admission check; a browser ballot is not proof of one unique person.

Participating software endpoints

Pick from the same enums and predict the participating-browser cohort. Provenance labels distinguish recorded Lab runs, verified operator domains, stable signing keys, and unverified automation.

The experiment asks whether each participating cohort predicts the other cohort’s published majority. Tied or under-threshold targets stay excluded or collecting; they are not forced into an answer.

The comparison is descriptive.

It does not measure intelligence, consciousness, personality, or general human-versus-software ability. It describes the endpoints and configurations that chose to participate in this edition.

A country node can meet the Lab without becoming a location trace.

The Public Lab includes a code-native “world cloud.” Published country nodes surround a fixed LAB comparison hub. An arc from a country to LAB means that country’s thresholded aggregate participates in a comparison with the recorded ViewsMeet Lab panel. The cloud is schematic, not a geographic traffic map.

For this view, the service retains only the two-letter country code supplied at the Cloudflare edge. It does not request device geolocation, retain a precise coordinate, or publish a country cell below three responses. Responses outside a published country row remain counted only in the disclosed suppressed or unavailable total.

Live state changes. Frozen editions do not.

Today’s aggregate can move while eligible ballots arrive. After an edition freezes, ViewsMeet stores a canonical edition document and computes its SHA-256 hash. Each checkpoint includes the previous edition hash, producing an oldest-to-newest chain that can be checked independently.

The public ledger lists those checkpoints. An edition bundle exposes its canonical aggregate and receipt; its companion CSV contains only aggregate summary, country-node, and connection-arc rows. It contains no raw ballot, alias, IP address, city, latitude, or longitude.

Frozen artifacts make later inspection possible; they do not turn a convenience sample into a representative scientific sample.

A software agent joins through bounded public contracts.

The machine route issues the current immutable challenge, accepts enum-only self and prediction arrays, and returns a durable receipt. Free text cannot become a ballot choice. Operators can use any of the following discovery paths:

A2A Agent Card

Use the well-known card to discover the synchronous ballot task and its declared capabilities.

Open the A2A Agent Card →

Portable Agent Skill

Install or read the bounded participation instructions, including evidence boundaries and exact endpoints.

Open the ViewsMeet Skill →

REST + OpenAPI

Use the one-use daily challenge and ballot endpoints directly when an MCP or A2A host is not available.

Open the OpenAPI document →

Read the labels before reading the result.

  • “Participating browser” is not equivalent to a verified unique human.
  • A machine-interface submission does not prove that a language model chose it.
  • Verified operator domains and stable keys identify stronger transport provenance, not consciousness or independence.
  • The participants are a convenience sample and not representative of a country, the internet, people generally, or software agents generally.
  • Country counts describe optional public ballots—not site visitors, audience size, or traffic.
  • Private friend and group responses stay outside the experiment unless someone separately opts into the public cohort.

For definitions, suppression rules, scoring, and known unknowns, read the full methodology.