We didn't guess.

Last verified: July 1, 2026. Companion to our data-trust page.

Every part of this platform maps to one of the best-replicated findings in how people actually become candidates. Below is the plain-language crosswalk: what we built, and the evidence it stands on.

The honest-numbers rule. The findings below come from independent field experiments, not from us. We will never quote our own program's win or completion rates as if they prove cause. Where we cite a number it's from the study named. It is not a claim about our results. (For example, the famous "85% cohort completion" figure is selection-confounded. We don't use it.)

The personal ask

Source: Gulzar & Khan, Review of Economic Studies 2024

What we built: nominate-a-neighbor and adopt-a-seat, plus the supporter referral spine. Why: being asked is the single best-established lever into candidacy. A doorstep encouragement conversation produced a roughly fivefold increase in actually filing (Gulzar & Khan, Review of Economic Studies 2024). Being asked is the modal path for two-thirds of officeholders. Women and LGBTQ+ people are systematically under-asked (Fox & Lawless 2010). Two refinements we bake in: prosocial framing ("your block needs") beat status framing, and public asks beat private.

Plan-making

Source: Nickerson & Rogers, Psychological Science 2010

What we built: the Plan and the weekly ritual. We also ship key dates + calendar export and the petition wizard. Why: a concrete when/where/how plan raised turnout where encouragement alone did nothing (Nickerson & Rogers, Psychological Science 2010); implementation intentions are among the most reliable effects in psychology (Gollwitzer & Sheeran 2006). The refinement: push specificity one level deeper: "Saturday at 10am by the farmers market."

No shame ever

Source: Milkman et al., Nature 2021

What we built: counters that never reset and badges that can't be lost. Every reminder carries a smaller "downshift." We add warm welcome-backs and "anchors slip" language. Why: tangible-reward gamification undermines motivation while verbal encouragement strengthens it (Deci meta-analysis); highlighting a broken streak backfires (Silverman & Barasch 2022); and the single best intervention in a 61,000-person study was a bonus for returning after a missed workout (Milkman, Nature 2021). Our reminder caps are evidence too: a second reminder adds little and carries real annoyance cost. So the preference center is required, not a nicety.

Relational with structure

Source: Cohen & Green 2023

What we built: supporter links that recruit through the candidate's own ties. Each one carries pledges and logs plus check-ins. Why: contact from someone you know outperforms cold outreach by an order of magnitude, but only under real accountability structure; the biggest structure-free scale-up found zero effect (Cohen & Green 2023). The structure is the active ingredient, so we never ship the bare version.

Structured cohorts + belonging

Source: Walton & Cohen, Science 2011/2023

What we built: the cohort gate and the brunch. We also built the board and the Queer Leaders page plus the wall of whys. Why: structured peer support persists where a bare assigned buddy does nothing (Cai & Szeidl 2018; the null buddy study by May 2006). "People like me belong here" interventions replicate across dozens of settings (Walton & Cohen, Science 2011/2023). They work where the environment offers real belonging, which the program is built to do.

Safety as infrastructure

Source: 2026 study of 215 out candidates

What we built: the campaign safety kit and the address honesty surfaces, plus the harassment-response plan. Why: a 2026 study of 215 out candidates found roughly two-thirds faced in-person harassment and a third faced online death threats, with fewer than one in ten able to afford security. Abuse measurably reduces willingness to serve (Brennan Center 2024), most for women and officials of color. This is a measured crisis with almost no tooling, so we built some.

How we measure

Only what people consent to. Every aggregate is floored at k≥5 (groups smaller than five are never shown). Our research carries no pixels and no third-party trackers. It has no fingerprinting either. Free-text (your words) is class-NEVER: it is never in any research payload. See the full data-trust page.

Our research layer is consented and anonymized by design. So this platform doesn't just use the evidence base. It extends it. It builds an outcome dataset where the public record is nearly blank. Want the aggregate briefing as that dataset grows? Ask to be added. Anonymized aggregates only, and you can stop any time. Questions? [email protected].