01
Hide the politics
A real government decision is presented without its title, sponsors, party positions, or recorded outcome.
A SmallProof experiment
Civic Mirror
Let your AI evaluate a real policy before it sees the politics.
It uses context you already share with your agent to predict how you would vote. You review the answer. Only then do you see the bill, party split, and result.
One real billParty labels hiddenNo context upload
The experiment
A prediction about you, not a decision for you.
Civic Mirror tests one narrow question: can an agent infer how someone would vote after considering a policy's effects and tradeoffs, without relying on party identity?
You remain the judge of the answer. A wrong or uncertain inference is evidence too.
02
Ask your agent
Using only context you authorize, it predicts support, oppose, or abstain and explains what could change the answer.
03
Review, then reveal
You judge whether the inference represents you. Only then do you see the bill and the recorded vote.
Why it matters
Elections bundle. Agents can separate.
A candidate stands in for thousands of policy decisions. Agents make it possible to examine those decisions one by one without asking people to become full-time policy analysts.
TodayOne reviewed inference
Experiment 001 reveals one bill and one House vote. It does not produce a personal alignment score or prove the larger governance thesis.
The longer testA record of representation
Across many decisions, reviewed judgments could show whether the people and parties someone supports actually represent their policy preferences.
The boundary
The agent can be wrong.
Civic Mirror does not assume the model knows you better than you know yourself. Its answer is a prediction, not objective truth or a diagnosis of political identity.
- You confirm, correct, or reject every inference.
- Abstain is valid when context or evidence is weak.
- Raw personal context stays with your existing agent.
Any future public aggregate would also need verified participants, standardized policy packets, model diversity, and privacy-preserving execution. This private preview does not claim those systems exist yet.
Experiment 001
Start with one policy.
The result stays private unless you choose to share it.