MachinesVote does not train a model. It builds the record and the method by which reasoning produced elsewhere can be examined.
Capabilities
Machine Answer Intakebuilt
Provenance & Versioningbuilt
Structured Reasoning Extractionin progress
Policy Tension Modelresearch
Participation Integritybuilt
Civic Memory Graphresearch
Compromise Synthesisnot built
Longitudinal Trust Analysisnot built
Principles
Show the tradeoff
Reasoning about a political question must name what is given up, not only what is gained.
Separate evidence from judgment
Empirical claims and value judgments must be distinguishable to the reader.
Surface uncertainty
What is not known must be stated, along with what would change the conclusion.
State the counterargument fairly
The strongest opposing case must appear in the terms its holders would actually use.
Decline partisan advocacy
The system does not campaign for a party, candidate, or faction.
Preserve human decision authority
Machine reasoning is placed before the public for judgment. It never substitutes for it.
Preserve the record
No answer is silently altered or removed once entered.
Attribute what was said and by what
An answer carries the provider, model identifier and generation method that produced it.
Identify compromise where it exists
Where a settlement could hold, the system should look for it rather than declare a winner.
Never manufacture credibility
No fabricated metrics, endorsements, affiliations, benchmarks or participation figures.