How Votr scores politicians, where the data comes from, and what we deliberately don’t publish.
Your alignment percentage with a politician is a comparison of two profiles: yours, built from the priorities you select during onboarding, and theirs, built from their voting record. Both are described in the same governed set of political dimensions (see below).
For each dimension on which both you and the politician have a meaningful position, we compare three things:
The percentage is the weighted similarity across every dimension where evidence exists. Dimensions without evidence don’t pull the score in either direction — they simply don’t contribute. The specific weightings are part of the scoring engine and aren’t published (see “What we publish” below).
Every politician and every voter is described in the same coordinate system: a governed registry of political dimensions spanning the major domains of public policy —
Each dimension carries a ratified definition with cited anchor sources, and every politician is rated on the same dimensions, with the same definitions, derived from the same empirical sources. This is what makes alignment comparable across politicians, across states, and across years.
The dimension set isn’t frozen: new dimensions enter through a governed discover-curate-ratify process as the political landscape shifts, and every stored score is stamped with the methodology version that produced it — so a score is always comparable to other scores from the same version, and never silently re-interpreted when the model evolves. Derived metrics like effectiveness and bipartisan rate surface as face stats below rather than contributing to the alignment number directly.
Alongside the headline alignment number, every politician card shows six independent stats — four about how well they representyou and two about how they operate universally.
When sources disagree, Votr trusts them in this fixed order:
When two sources disagree, the higher-trust one wins and the lower-trust record is quarantined — visible to internal audit but not surfaced as evidence.
Your political profile is computed on your device. Votr’s server never receives your tag selections, your dealbreaker flags, your alignment scores, or any vector representation of your views. The server sends you raw politician data; your browser compares it against your local profile and renders the alignment number locally.
We call this Privacy by Impossibility — the server can’t leak data it never sees, because the architecture makes seeing it impossible. The trade-off is that some features (cross-device sync, shared cart export) require end-to-end encryption when they ship; the math runs on your devices, not ours.
We publish what the score means and where it comes from: this methodology description, the dimensional model, the data trust hierarchy, the privacy model, and — most importantly — per-score receipts: every alignment number can be traced to the specific votes, sponsorships, and recorded actions behind it, on the page where it appears.
What we deliberately don’t publish: the scoring engine itself — the algorithms in implementable detail, the numerical weights, the formula coefficients, and the calibration data. The engine is a proprietary trade secret while Votr operates.
The reason isn’t secrecy for its own sake. Once the specific weights are public, sufficiently-funded actors can gradient-descend their messaging and voting profile against the score, gaming the number without changing their behavior. Keeping the engine private preserves the score’s usefulness as a signal of real alignment, not as a target to be optimized against.
The escrow guarantee: the engine’s full source is held in escrow. If Votr is ever captured away from its mission or ceases to operate, a dead-man’s switch releases the entire engine publicly under AGPL v3. Secrecy in life, openness on death — you never have to trust that the engine will outlive our good intentions.
Every change to the methodology is recorded in the public scoring changelog.