A modeled public · United States + world
Model how public reaction could move.
Pheme is a simulation—not a poll. Ask a question, test a scenario, and inspect how modeled reactions vary by place.
Named public demographic and geographic inputs.
A question or explicitly hypothetical scenario.
Estimated patterns, shown with sources and limits.
Process diagram only. It shows no observed opinion, geography, or result.
Choose your instrument
Four ways into the signal.
Start with a task. Method and supporting detail stay close without blocking the work.
Put a question to the model
Compare modeled choices and coalitions across a chosen place.
Ask a question → 02 · SimulateWatch a scenario unfold
Follow a cinematic, deterministic reaction across the U.S. or world.
Choose a scenario → 03 · ExploreInspect the geography
Move between maps, rankings, profiles, sources, and tables.
Open Explore → 04 · EvidenceRead the source first
Open original reporting, inspect claims, then model a response.
Open the news desk →Observed versus modeled
A model earns trust by showing the miss.
Pheme publishes its committed validation set, source links, average error, and worst result. Direction can be useful while an exact percentage is still wrong. This is a simulation, not a poll.
Inspect every benchmark and source →Loading the committed scorecard…
Issue count, run count, holdout status, and artifact identity loading…
Current brief
Read the source. Then inspect the signal.
Loading the latest sourced stories…
A question, sourced story, bill, or hypothetical scenario.
Deterministic geography plus bounded synthetic-persona reasoning.
Result, sources, version, limitations, and validation together.