Methodology

What this is, and what it isn’t

This is not a poll, and its results are not representative of any population. It’s a self-selected online convenience sample: you found this site somehow — a link, a post, a search — and chose to click through and rate policies. People who click through are not a random sample of the public, and no amount of demographic data collected afterward changes that.

Collecting age range, state, party affiliation, and presidential vote history lets us show how the sample splits along those lines — it does not make the sample representative, and it is never used to re-weight the results toward some estimate of “the public.” We will never publish a number phrased as “X% of Americans.” If you see this site’s numbers described that way anywhere, that description is wrong.

Every figure below is published with its sample size, the date range it covers, and the quality filter applied to it — check those three things before drawing any conclusion from a number.

The instrument

You’re shown a shuffled set of policy actions, stripped of the administration, date, and any other detail that would let you guess who did it. Each one, you rate on a single five-point scale from strongly oppose to strongly support — nothing else is blended into that scale.

Each policy also carries two hidden axis loadings, fixed before any item was written:

Your position on each axis is a loading-weighted average of your ratings — the exact formula is fixed and versioned, see scoring_version below. Combining the two axes with an intensity split (near the center vs. far from it) produces one of 8 archetypes — a label like “Bold Builder” or “Steadfast Localist” — which is what you see on your results page and what the crosstabs below are broken out by.

If you rated too few items to place you meaningfully on an axis, that axis is marked “undetermined” and defaults to the center rather than silently becoming a fake data point — it’s excluded from every axis-level average below.

Where the deck comes from

Every item cites a primary source — an executive order number, a bill number, or a Federal Register citation, linked from the item’s expanded detail. No paraphrased news coverage. Items are written to a neutral, parallel construction and stripped of proper nouns and signature phrasing that would identify an administration — some residual recognizability is tolerated (a hidden test can’t erase 40+ years of memory), but the goal is to remove avoidable tells, not guarantee unidentifiability. The deck is checked, before publication, for rough balance in item counts by party and administration, and for whether either axis’s loadings skew toward one party — an unbalanced deck would make an axis a partisan label with extra steps rather than a real dimension. The current deck is a 10-item seed set spanning both parties and multiple administrations, not the full 40–60 item deck; treat any numbers below as preliminary until the deck is complete.

Editorial calls are made by the project owner; there is no reader panel and no pre-launch vote on individual items. That’s a deliberate tradeoff, not an oversight — see “Editorial bias” below for how it’s checked instead.

Quality filtering

Every submission gets an advisory quality score (1 − 0.2 × number of flags, clamped to [0, 1]), based on signals like rating too fast to have plausibly read the item, straight-lining the same answer across many consecutive items, or a suspiciously monotonic answer pattern by presentation order. Nothing is rejected at submission time — a flagged response is still recorded in full — but figures published here exclude any submission scoring below 0.6 (two or more flags). That threshold is a provisional default, not a value derived from any external validation study; it will be revisited as real submission volume accumulates. Every table below states this threshold explicitly rather than filtering silently.

Editorial bias: how it’s checked, not just asserted

Because there’s no pre-launch reader panel, bias-checking happens after launch, against real submission data, on a manual, non-automated cadence (informally, roughly monthly): does an item’s support rating split by declared party more sharply than the rest of the deck does? An item that stands out gets reworded — which retires its old identifier and starts fresh data collection under a new one, so old and new ratings are never silently merged — or down-weighted. This is a real, ongoing process with a stated method, not a promise with no mechanism behind it; it is also manual and informal rather than an automated dashboard, and its specific statistical threshold reuses the same balance check applied to the deck at build time rather than a bespoke, independently validated one.

Privacy and geography

Political opinion combined with demographics is treated as sensitive data. Submission (not reading) is blocked from the EU, EEA, and UK, where it would be regulated more strictly than this project is built to support. See the privacy policy for the full data-handling explanation, including how identifiers are made unlinkable and how to request deletion.

What we’re not claiming

If any of the above changes materially, this page is updated to say so — not quietly, and not after the fact.