Empowered Vote · Volunteer Briefing · October 5, 2026
Every number on this page comes from the live database, verified git history, or our product analytics, not from estimates. We build tools for shared facts; our progress report should be one. Each briefing is written fresh and kept at its own dated address, so you can read what we said before and check it against what we say now.
Six cities in three states were researched end to end in four days — Charlotte; Bradenton, Miami and Tallahassee; Duluth and Saint Paul. Forty-seven seated members, thirty-five topics each, 1,645 researched rows. Thirty-seven of those rows propose an actual position, and every one of the thirty-seven is in front of a human being rather than in front of a voter. Nothing in this programme publishes itself, and nothing has: the stance total in the band above has not moved since September 27, and that is the honest reading of the week. We would rather report a research week as a research week than let a headline number drift upward on work nobody has checked.
Nearly half of the work is proof that a city is not allowed to act. 792 of the 1,645 rows are scope blanks, because a state can reserve a subject to itself and most do. That is a finding, not a failure — and it is why Minnesota was chosen third, to test the opposite case. It held: four ladders came alive there that neither Carolina nor Florida could reach. The detail is below.
The largest finding of the week was an evidence channel nobody had used. A measure that passes unanimously, by voice, records no division — so the vote says nothing about who wanted it, and our tools read that silence as absence. Asking instead who wrote it seated four Duluth members at once, off one resolution. Three tools were blind to it at the same moment: the divided-vote scan, the member vote lists, and every news sweep we run — not one article in any of the nine press corpora names a sponsor. When a measure passes without division, stop asking how people voted and ask who wrote it.
And one slice of our research was certified reliable by measurement rather than by assertion. Three AI coders read the same 57 state-level, record-based stances against human labels written blind, and agreed closely enough to clear the bar — with no severe disagreements at all. The same 57 items had failed that test before the rulebook was written. Figures below.
Each update is a dated snapshot. We do not edit one after publishing it.
This page is the current briefing, and it always lives at empowered.vote/briefing. Every edition also keeps a permanent address of its own, so a number you read here can be checked against the date it was published on. Each update is written from scratch: work that runs for a month stays the headline, but the sentences around it are rewritten every time. We never edit a published edition. When an earlier one got something wrong, the next one says what it corrects.
What the edition of September 29 left out. It headlined four states seated and named North Dakota, Kentucky, Kansas and South Dakota. A fifth had been seated in the small hours of that same morning and went unmentioned: 188 Missouri General Assembly seats at 02:52 UTC and the City of St. Louis’s 22 at 06:16, both before the edition was written. Nothing it printed was wrong; it simply did not look. Missouri now holds 155 of 163 House seats, 33 of 34 Senate seats and all five statewide executives.
It also set itself one test, and the test came back empty. It said the trivia audit “landed too late in this window to have moved” the completion rate, and that “the next pull is the one to read.” This is that pull: 12 people finished a trivia game in the quarter out of 111 who arrived — the same twelve, out of the same hundred and eleven. Six days is far too short to conclude from. We record the reading rather than drop the promise.
Open to everyone, no login. Evidence-based, engaging, anti-partisan.
| Feature | Status | Journey | Where it stands | What's next |
|---|---|---|---|---|
| Civic Trivia Championships | Live | alpha |
Where it standsA player who noticed where the right answer sits could have beaten this quiz. Across 117 freshly generated questions the correct answer landed on option C 41% of the time, and on 61% of the questions whose options are numbers. The cause was ours: the code sorted numeric options into order after generation, and models write distractors that bracket the true value — two below, one above — so the answer kept ranking third. Bank live today: 3,277 questions across 43 collections. | What’s nextThe fix inverts the order: the position is drawn from the question’s own identifier first, the generator is told where to put the answer, and the old code now verifies instead of imposes. A gate fails a batch above 35%. Those 117 questions sit in two collections that are switched off until they pass. The content audit did not advance: one collection still needs writing rather than reading, and two remain under the expiring-content floor. |
| Empowered Essentials | Live Alpha | alpha |
Where it standsZIP-to-representatives with funding, votes and stances. The front door lost a search box and gained thirteen cities. There were two ways in, and search-by-name answered worse than the address path beside it; it is gone, and thirteen communities we already serve now have a chip to tap. Indianapolis is keyed on the county FIPS code, because the city and the county are one government there and a city FIPS finds nothing. | What’s nextBanners: St. Louis takes the Arch, so Missouri moved to the Ozarks; Indiana gave up the Indianapolis skyline for Indiana Dunes. No photograph may appear twice on one page, so a city joining pushes its state to a new subject. Still owed, unchanged for a second cycle: portraits for the 70 Puerto Rico alcaldes without one. |
| Empowered Compass | Live | alpha |
Where it standsEvery one of the 60 topics a voter can answer now has a written annex — what each rung means, where the line between two rungs falls, and the judgments already made. All 60 were drafted or revised this cycle, out of a library of 64. No change under the voter: Season 2 is still open, and we check for a cutover every cycle because the numbers do not announce one. | What’s nextSeason 3 is being assembled in the open and now holds 61 staged topics. The city work found real defects in the local-tier ladders — rungs no city in three states could occupy — written up as Season 3 input rather than patched in place. Then reliability measurement extends past state records to the other evidence classes. |
| Treasury Tracker | Live | alpha |
Where it stands8,144 governments, unchanged. No commits at all this cycle — the speed work landed on September 23 and nothing has followed it. Arrivals rose anyway, 148 people in the quarter to 169. | What’s nextThe debt the last edition named is still outstanding: we measured the slow call against the fast one, 3.05 MB against 7.4 KB, but never the five-run before-and-after on the real page. A faster call is not a faster page until the page is timed, so the speed claim stays a claim about one call. |
| Read & Rank | In Development | build |
Where it standsA blind taste test for candidates: the question, not the race, is the unit of comparison. Its first commits in four cycles, and they are a volunteer’s UX redesign prototype rather than product code. Still gated on the stance pool — and this cycle’s research went into rows awaiting review, not into the pool. | What’s nextThe prototype shipped with invented quotes under the names of real members of Congress, disclosed in a README the rendered page did not carry. Fixed the same day: fictional stand-ins, and the disclosure moved onto the screen itself. On a platform built on provenance, placeholder text must never sit under a real person’s name — design documents included. |
| Fallacy Finders | In Development | build |
Where it standsv1.1 is 4 of 6 phases done and built the Fallacy Sleuth: curation-time AI proposing timecoded fallacy tags, with names stripped before the model sees a word and every clip re-run with the speakers swapped. Quiet for a fifth cycle. | What’s nextIts dependency moved instead. On the Record attempted two US House floor days and its quality gate refused both — a full legislative day came back as 13 segments and then 9. That refusal is the system working, and it is also the honest status of rights-clean clip sourcing at volume: not there yet. |
Connected accounts, one human one voice. Parked during the election push; resumes after November.
| Feature | Status | Journey | Where it stands | What's next |
|---|---|---|---|---|
| Civic Spaces | Live | alpha |
Where it standsGeo-scoped communities, neighborhood to federal. A third busy week in a parked pillar: the banner work became a shared library rather than a per-screen asset, the location banner became compact, and the dashboard got its layout, post cards, events and weather. | What’s nextThe offline place-name table from last cycle is merged into that stack. The pillar stays parked until after November, and this is groundwork being laid early rather than a change of plan. |
| Empowered Quests | Live | alpha |
Where it standsFact-verification quests with consensus scoring; verified facts already feed Essentials. Red gem quests unlock at VR 90+. No commits this cycle. | What’s nextv1.6 planning. |
| Focused Communities | Paused | build |
Where it standsv1.0 shipped April 17: per-topic forums wired to the Compass. Waiting on the post-election Connect pivot. | What’s nextv1.1: badge ratification and argument mapping. |
| Empowered Listening | Paused | build |
Where it standsStructured debate infrastructure; 5 of 6 phases done — timers, transcription, audience notes. | What’s nextPhase 6: voting and badges. Then Common Grounds can build on it. |
| Emparks | Design | design |
Where it standsPlatform-wide annotation system, five mark types. Full design doc; no build yet. | What’s nextAwaits Connect-pillar capacity. |
| Common Grounds | Design | design |
Where it standsAdversarial-collaboration mediation. Full design doc; depends on Empowered Listening and Fallacy Finders arbiters. | What’s nextBuilds after its dependencies land. |
The shared spine every app plugs into.
Auth, tiers, XP and gems, politician database, TIGER geofencing, elections. The effort went into research rather than seating, but 240 dated terms landed: Missouri’s legislature and executive, the City of St. Louis, St. Louis County, and Indianapolis and Marion County merged into a single government on the precedent Nashville set here last month — a 25-seat city-county council, 20 of whose seats were newly dated. 10,959 districts are mapped, up 27. 121,178 legislative votes and 19,622 bills on record.
Speaker-attributed, searchable transcripts of public meetings, synced to the original video. It reached a state legislature this cycle: the Indiana General Assembly’s own video system resolves now, and a Senate Judiciary Committee hearing is published at 350 segments and 33 speakers. The dataset reads 172 meetings, 21,452 transcript segments and 1,409 identified speakers, including LA City Council with 7,200 recorded votes. The standing caution is repeated rather than retired: segments carry no creation timestamp, so dated meetings are the evidence here, and the two drops earlier this year are still unexplained. Built by volunteer Chris Andrews.
The Compass places a politician on a ladder from 1 to 5. Until now, the claim that those placements are consistent rested on the people making them. It is now measured. Three AI coders read the same 57 state-level stances that rest on a legislative record, and their answers were compared with human labels written at a blind desk, where the labeller never sees what any model said. The two agreement statistics came back at 0.830 and 0.932, the third test at 0.904, and the count of severe disagreements — two readers more than one rung apart — at zero. That clears the bar; the stratum was certified on October 3 by volunteer Chris Andrews.
Two controls are why this is worth printing. Eight of the 57 items had themselves been used as examples while the rulebook was written, which is a quiet way to grade your own homework; removing them leaves 54 items at 0.821 and 0.928, still certified. And the identical 57 items, coded before five rulings were added, scored 0.741 and did not pass — so the gain belongs to the rules, not the models, because the rulebook is the only thing that changed. One stratum of one evidence class is certified, and that is all. News-sourced and statement-sourced evidence are not, nor is the local tier the city work above lives in.
The invisible work. This is what "evidence-based" costs, and we paid it.
36,438 stance positions across 4,196 politicians and 64 topics, every one carrying reasoning and cited sources; 1,846 of the 2026 candidates already have some. One figure keeps its date said out loud: 2,722 sources were independently re-checked in a sweep that finished in April 2026, and the number has not moved since, because every row in it was written then. It is a dated achievement, not a running total.
| Coverage area | Politicians | Stances |
|---|---|---|
| US Congress + national | 556 | 9,494 |
| 2026 candidates & discovered names (no seat) | 933 | 6,032 |
| California (state + local) | 372 | 4,767 |
| Massachusetts (state + local) | 302 | 3,355 |
| Maryland (state + local) | 188 | 2,228 |
| Texas (state + local) | 270 | 2,073 |
| Maine (state + local) | 186 | 1,794 |
| Virginia (state + local) | 166 | 900 |
| Oregon (state + local + county) | 154 | 824 |
| Utah (state + city + county) | 57 | 650 |
| Wisconsin (state + county + local) | 208 | 628 |
| Washington (state + county + city) | 161 | 489 |
| Arizona (state + local) | 129 | 436 |
| North Carolina (state + local) | 124 | 406 |
| Tennessee (state + local) | 119 | 231 |
| 38 more jurisdictions (incl. Florida and DC) | 271 | 2,131 |
No answers were written this cycle, and fourteen of the sixteen rows are identical to the last edition’s. The two that moved are a matched pair: one person left the no-seat row for a jurisdiction, carrying six answers into the remainder. The rows still sum exactly to the headline totals. That is the strongest form of the check this table can give. The sum alone catches a missing row but not a row filed in the wrong place, since a misattribution moves a count between buckets and leaves the total intact. Fourteen rows reproducing to the number says the resolution order did not shift underneath us.
How a stance gets seated. A member is placed on a topic because of a bill they sponsored, read in its session-law text — not a sentence someone else wrote about them. Each of the 60 topics now carries a written annex saying where the boundary between two rungs falls, and that rulebook is what the measurement above was measuring.
Treasury Tracker’s database holds budget data for 8,144 government entities across all 50 states plus DC, spanning fiscal years 1976 to 2028:
A city is not a small state, and the first question is never what someone believes. It is whether the body they sit on may decide the thing at all. North Carolina and Florida reserve firearms, municipal wages and rent regulation to themselves, so 522 of those two tranches’ 1,015 rows are proof of a limit rather than a position. Minnesota reserves fewer — Saint Paul has a rent stabilisation ordinance, a municipal minimum wage, a chapter limiting city help with federal immigration enforcement, and ranked voting. But it reserves cannabis, which Florida does not, so a finding copied from one state to the next would have been wrong in both directions. The scope review runs before any research and is itself cited, to the statute and the municipal code chapter.
Then the sources have to be able to carry the claim, and ours sometimes could not. Three Saint Paul members were cited to one page: a meeting agenda that mentions two of them zero times and carries no roll call at all. The citation checker was right to refuse all three. The page that could carry the claim was one link away, and all three rows now verify on three sources of three with no value and no reasoning changed. When a citation fails, ask first whether the page could ever have carried the claim.
And a thin result is a claim about your tools until you have checked them. Re-mining the same cities with a link extractor that assumes no one publisher’s layout took Charlotte from three seated rows to five, Florida from none to two, and Florida’s attributed passages from 22 to 122. One member’s own press sweep returned 700 files of which nine named him, his surname being a common one; every passage finally used came from other members’ corpora. Price the next city off the second set of numbers.
The third leg, alongside votes and stances. The platform holds 29.1 million individual campaign-finance contributions totaling $8.6 billion, covering 2,248 federal politicians and candidates across 24 election cycles (1980 to 2026). Every record traces to its original FEC filing, and superseded lines from amended filings are retired individually rather than counted twice. Unchanged this cycle: the widened matching and the 33 disputed surname links were last cycle’s work, and nothing has arrived since.
Every state has at least one researched official, and all 50 congressional delegations are covered nationally. Shading shows the depth of state and local research. This map did not move, and the city work above is why: a tile shades only when an answer is published, and none was. The next edition is where Charlotte, Miami and Saint Paul would show — if a human approves them. Puerto Rico is an outlined tile: 161 seats seeded, no stances yet.
Hover any state for its exact count. The seeded tier is mostly governors and statewide officers; deeper tiers add legislatures, counties and cities.
Who actually uses what we build — measured, not guessed.
673 people used the platform in the last 90 days, across every app, over 1,644 sessions. What they do once they arrive is good — sessions average six minutes sixteen and 7 in 10 don’t bounce — so the constraint is arrival, not engagement. 105 of the 337 who reached the main site clicked through into an app: 31% in the quarter, the third pull running at that share and each on a larger base.
Read every figure here as an order of magnitude, not a trend line. At this volume a couple of people either side is the whole swing, and the window slides, so an app can “fall” because a busy week in July rolled out of it. Over seven pulls the main site reads 306, 316, 312, 297, 300, 314, 337. The hostname map was re-run first and found no new host, which is what makes this pull readable against the last six.
Everyone who arrived, against the people who did the one thing that app exists for. Bars are the 90-day rate; counts are people.
Two of these are start problems, not finish problems. Only 19 people started a trivia game in a quarter, on 111 arrivals; 12 finished. Read & Rank is the same shape: 19 races started, 8 ballots revealed. The Compass is the opposite, and the healthiest funnel we have: 123 of its 138 visitors opened the quiz and 41 saw it through. The trivia row is unchanged from last time, to the person — see the archive note.
Unique visitors per app, 30 days and the full quarter. Each app is counted once across every hostname that serves it — Treasury drew traffic on four this quarter, the main site on two.
Treasury Tracker is still the second most-visited thing we run, and it rose furthest this pull — 148 to 169 in the quarter — in a cycle with no commits at all. Civic Spaces is absent from this table, and that is a measurement gap rather than a result: it serves no traffic our analytics can see.
An earlier edition said seven people in a quarter opened any of Treasury Tracker, as 5% of 152 visitors. That rate was not sound and it stays withdrawn. The events were fine; the denominator was not — Treasury pushed its in-app navigation into the address bar without recording a page view, so every visitor was filed against the bare domain. The direction was wrong too: this quarter those few produced 133 category drills, 93 entity selections and 71 line-item views.
The fix has recorded since August 16, and the two ways of counting still disagree by more than four times. 72 people reached a budget address; only 17 fired any past-the-front-page event. The split explains it: 55 landed on our own budget, one click from the donate page — a transparency visit, which selects nothing and so fires nothing. That leaves 19 on some other government’s budget, and that half moved for the first time in three pulls, 16 to 19, while the donate half went 47 to 55. Treasury stays unmeasured rather than low until the half that is not ours outgrows the donate traffic. A low count is a claim about instrumentation until you have read the code path.
How much further we have to go, honestly.
Review the 37. A proposed position nobody has approved helps no voter, and the city programme makes more of them every week it runs: the review queue is the bottleneck now, not the research. Then the state legislators, where the evidence per row is far better than in any city — 2,551 seated people across 16 legislatures are in scope. Finish the 2026 candidate waves, write the one trivia collection an audit cannot fix, and time the Treasury page properly. Success in November is one thing: any voter in a served area can look up every name on their ballot and follow the money behind it.
Reactivate the parked pillar: Empowered Listening phase 6, Focused Communities v1.1, Civic Spaces v4.0 — whose shell, shared theme, banner library and dashboard are groundwork done early. This is where the platform stops being a reference tool and becomes a community, and where our sustaining-donor model lives.
Designed but not built: the Gems stipend economy, Awareness Exchange, Badges, Symposium tournaments, Emparks, Common Grounds, and tools for citizens to run for office on the strength of their solutions. Roughly a third of the founding vision, waiting on the first two pillars.
A volunteer-built platform, visible in the commit history.
An interactive Purpose mind map walks the whole vision — the thesis, and the Why, How, Who and What — with a short explainer on every concept and pillar feature.