Torus interviews voters at scale, models the narratives that move them, and turns a balkanized electorate into a working map — so your campaign can speak to the coalition you actually have.
Static polling and a quarterly memo aren't the cycle democracy actually runs on. Torus closes the loop — each stage feeds the next, building feedback into the loop.
Conduct 15-30 minute interviews with stratified samples of the target electorate — in voters' own words, not closed-end checkboxes.
Train a campaign-specific narrative model on the transcripts plus environmental context. Surface the latent segments and the language that binds them.
Generate dynamic message recommendations by segment, channel, and moment — ranked by predicted persuasion lift on the voters who matter.
Field response, ad recall, and follow-up interviews close the loop. The model re-weights as needed — the brief you get in October is not the brief from June.
Torus replaces the polling memo, the segmentation deck, and the messaging guide with a living workspace your strategy team checks every morning.
A living portrait of the electorate — multiple thematic narrative segments, the language that defines them, and how to approach them.
Segment-tuned message recommendations with predicted lift, sample copy, and the voter quotes that anchor them — in the language your voters actually use.
Track how a deployed message lands — in ad recall, in social listening and in the field — and the model re-weights itself before your next buy.
Torus identifies the latent narratives voters carry — the stories they tell themselves about the country and their place in it — and surfaces where seemingly disparate voters share belief, priority, and emotional register.
A single survey tells you where a voter stands today. Torus follows the same voters across cycles — modeling not just position, but velocity, bearing, and the events that bend the line. The electorate is a system in motion; we treat it as one.
How fast a voter is moving through narrative space. Rapid shifts mark a persuadability window — the moment to enter the conversation. Slow drift means stable cohort behavior; don't waste the buy.
Two voters in the same segment today can be heading in opposite directions. Bearing forecasts where they'll be next cycle — so the brief reflects who they are becoming, not just who they were.
Events bend the line — a layoff, a school board fight, a hospital closure, a wedge ad. We model both the event and the magnitude of response, so you see causation, not just correlation.
Strategists, organizers, and creative teams all live in the same view — segments, messages, lift, and the voter language behind every recommendation.
Closed-end polling tells you what people will agree to. Torus is built on what they actually say — long-form interviews, contextual data, and a model that treats narrative as the unit of analysis, not the issue.
15-30 minute conversations sampled to district-level quotas. We start from voter language, not pollster categories.
An LLM trained on transcripts plus environmental data: media diet, regional economy, ballot context. Each campaign gets its own model, not a shared one.
Message candidates are scored against the model, then validated against held-out interviews and in-field ad recall waves before recommendation.
Every recommendation links to the voter quotes and segment evidence it rests on. Your team can audit the chain back to the human voice.
A campaign is only as honest as the data underneath it. Torus runs on dedicated U.S.‑based hardware, encrypts every byte in motion and at rest, and treats role-bounded access & full audit trails as table stakes — so the voter is protected, the data is contained, and every recommendation can be traced back to its source.
Torus runs on dedicated hardware in U.S. data centers (OVH Hillsboro, OR) — not multi-tenant cloud. Compute and storage sit on machines controlled exclusively by platform operators, eliminating noisy-neighbor and shared-hypervisor risk.
TLS 1.2+ on every API, UI session, and inter-service hop. Voter PII, interview transcripts, and extracted propositions sit on encrypted database volumes. Third-party LLMs are configured for zero data retention; an on-premises model is architecturally supported as fallback.
Per-account authentication with role-based access control — platform administrators, operators, consultants, and campaign staff each see only authorized views. The licensed IDGraph is reachable only through an API peering layer; no direct database, shell, or codebase access leaves the core operator perimeter.
Industry standards for data security. Exceedingly high standards for data hygiene — because a sentiment engine is only worth what its sources are worth.
Tell us about your race. We'll send a sample sentiment brief and walk you through the workspace.