Ballots and Run Records

After discussion, AnthroSim asks every participant for a final position, rationale, and confidence. The saved run preserves each ballot and computes the room tally from those records.

Example excerpt from the JSON export:

{
  "votes": {
    "final": [
      {
        "participant": "Maya Torres",
        "position": "support",
        "rationale": "The packet answers the ridership question, but feeder-bus funding still needs a commitment.",
        "confidence": 0.62
      }
    ]
  },
  "source_lookup_events": [
    {
      "query": "Is feeder-bus funding committed?",
      "provenance": "uploaded_source",
      "consumed_text": "Ridership projections assume expanded feeder-bus service. Funding is subject to a later operating agreement."
    }
  ]
}

The operator can review:

• Transcript: every public turn with its simulated time and speaker
• Private traces: structured records of decisions to speak, wait, or update a position
• Participation: spoken turns, silent thoughts, passes, and moderator invitations
• Evidence: each lookup, consumed passage, provenance, citations, and support-check diagnostics
• Ballots: every final position with its stated reason and confidence

The downloadable JSON also includes the generated profiles, complete run configuration, models used, and orchestration version.

Reading the result

The final tally is exact for that saved synthetic room. Report it with the audience definition and run settings. A second run can differ because model generation is stochastic.

The tally is not a population estimate. Demographic splits from one synthetic room do not establish how a real subgroup will vote, buy, or decide a case. Use the record to identify questions for real respondents.

Back to: Audience construction · Conversation simulation