Reading Their Own Words

Most people asking for your vote have never held office, so there is no roll call to read. The only hard evidence of what they think is what they have said in public, at length. This is the machinery that reads those hours, one transcript at a time, against a standard published in advance — with a verbatim quote behind every judgement and a link back to the tape.

One transcript fans out into one independent assessor per qualification area, each producing a pair of numbers, and every published number walks back down to the sentence and the timecode that produced it.

What it does

The movement keeps a qualification catalog — eighteen named areas, from anti-corruption and constitutional competence to economics, national security and loyalty to the USA. Each is paired one-to-one with its own assessor: a prompt, a calibration file, and a rubric of forty-five to sixty-five leaves, each with a ten-rung ladder describing what an answer at that depth sounds like. The rubrics are published before anybody is read against them, and they never contain a person's name.

The unit of work is small on purpose: one transcript times one qualification is one run. A two-hour interview fans out into eighteen independent runs. When one fails, the result is a named gap in that one area — not a hole in the video and not a quiet zero everywhere.

Four floor rules bind every run:

• Quote or silence. No verbatim quote from the transcript, no result.
• This transcript only. Nothing known about the person from elsewhere enters the reading.
• Their own words, never the interviewer's. A sharp question from the host is the host's.
• Silence produces nothing, not a low result. Writing a number for something a person never said is a false statement about a named living person.

Every leaf produces two values that are never averaged together: how the person did, and how much this transcript taught us. A strong reading on a thin transcript stays both — a high percentage beside a low known percentage — and is never corrected down to look reasonable.

The specifics a rung names — an instrument, a statute, a dated episode — fix the level of reasoning being asked for, never words the person had to say. Before denying a rung for a missing specific, the assessor runs the substitution test: would the same depth of answer, about a different instance of the same kind, satisfy this rung? If yes the rung is met, and the run records which abstract class it credited — a mechanism, an anchor, who pays, or what would change their mind. Understating a real person is exactly as serious a defect as overstating one.

Then it walks back down. Every published figure drills to a page for one person, one video, one area, where the transcript renders as timestamped statements per speaker. The passages that moved the number are highlighted in place, with an icon and a label rather than colour alone. Clicking one opens the leaf, the rationale in the assessor's own words, and a control that seeks the video to that second.

The goal it serves

Most of the people worth electing have never held office, so there is no roll call to do arithmetic over. Judging them by party label is the habit that produced the current Congress. What is left is what they have said in public, at length — hundreds of hours per serious candidate, far past what any volunteer will sit through. This is how a movement with no staff reads that pile anyway, against a standard published in advance and applied identically to everyone.

The part that matters is not the number. It is that the number is walkable. A citizen who thinks a judgement is wrong can click it, read the exact sentence, watch that second of the recording, see which rung it was matched to, and disagree with something specific. That is the difference between verifying a claim and trusting whoever made it — and a movement that asks people to stop taking institutions on faith has no business asking for its own.

What keeps it honest

• The gaps are published, not hidden. Leaves nobody touched, transcripts held but not yet read, and videos with no transcript at all are listed on the page, each with the action that would close it. A quote the system cannot locate in the transcript is reported as a count, never printed as words.
• The two numbers travel together. An award percentage printed without its known percentage is a defect, and neither blank state is ever rendered as 0%.
• Nothing crosses out of this stream. What a person said feeds the Flamingo Award and nothing else. The awards computed from votes cast in office never see a transcript.
• The assessor does not decide the outcome. It reads and quotes. The map from rungs to a percentage, and the threshold at which an award is given at all, sit in a versioned policy file anyone can read — and moving that threshold to make a distribution look better is forbidden in writing.
• It has caught itself. A sweep of 10,180 rungs across 807 videos found the top three rungs reached seventeen times and the highest never — evidence the rubrics were being read as an answer key. That is a fact about the reading, not about the people, and it is why the substitution test exists.

What it does not claim: that reading somebody's words tells you how they will vote.

Works with

• Videos, Transcripts And Provenance — supplies the timed transcript this reading runs over.
• The Qualification Areas — the eighteen areas and their rubrics, one assessor each.
• The Politician Record Page — where these readings surface, beside everything else known about the person.
• The Rosters — supplies the people to read, including challengers with no votes to examine.
• Your Seats And The Seat Map — narrows the stack down to the people on your own ballot.

Where to go next

• Qualifications — the eighteen areas, and why each one is asked about.
• Read the rubrics — the published standards, rung by rung.
• What feeds the Flamingo Award — where a reading of somebody's words ends up.
• What the AI never decides — the line it is not allowed to cross.
• Take Action — including adding a video to a record, which is how the pile grows.