Vol. 01Research desk

Corrections for the age of spectacle

The Dehyper

Strip the spectacle. Keep the facts.

House method

How a story gets dehyped

The pipeline is research first, prose second. AI can draft. It cannot invent a paper, a quote, or a number. Dehypes start from one viral object. Originals start from a subject. Same citation rules, no claim board, no hype index.

  1. 01

    Ingest the spectacle

    We start from the thing people actually saw: a headline, a keynote, a YouTube thumbnail, a founder thread. The agent pulls the readable text, the title, and the publisher. If it is a video, we take the page, the description, and a transcript when one exists.

  2. 02

    Split claims from weather

    A mood is not a claim. “AI is getting scary” cannot be falsified. “Model X passed the bar exam, therefore it is a lawyer” can. We extract the smallest statements that could be true or false.

  3. 03

    Go get the boring documents

    Press releases lose to papers. Papers lose to the actual table. Marketing decks lose to the limitations section. Survey “80% of experts agree” loses to the question they were asked. The agent searches for primary sources. The desk refuses invented citations.

  4. 04

    Score the hype, not the technology

    A high hype index does not mean the underlying work is fake. It means the distance between the evidence and the story being told is large. GPT-class models are real. “PhD-level employee” is a slogan.

  5. 05

    Write it small enough to be true

    The article has to name the original, quote the claim, show the record, and list every URL a skeptical reader needs. Drafts stay unpublished until the claim board and the source list can survive that read.

  6. 06

    Put the file on the first screen

    Readers often leave after a few paragraphs. Every piece opens with the viral claim, the smaller true thing, and a thirty-second file: what holds, what was inflated, and what is still open. The rest of the article is the breakdown. Important stuff first. Dossiers later.

How we tag stories

Every filed piece carries two kinds of labels. The verdict is the stamp on the viral story as a whole: what did the thumbnail sell, and how far is that from the record? The claim board scores each specific sentence. A piece can be missing context overall while individual claims are true, false, or unverified.

Article verdicts

One per piece. Shown on the article header and index cards.

Overstated

A real event or paper got inflated into a bigger story than the evidence supports.

False

The central claim does not survive contact with the record.

Missing context

The facts are not invented, but the framing leaves out the part that changes the meaning.

Narrowly true

Something happened. It is smaller, more local, or more conditional than the headline.

Unknowable

The claim is a forecast or a vibe. There is not enough evidence to cash it out.

Claim findings

One per row on the claim board. Each is the smallest falsifiable statement we could extract, scored against primary sources.

Holds

The claim checks out against primary sources. The hype may still be wrong overall.

Overstated

Directionally related to reality, but scaled up in timeline, scope, or certainty.

False

Falsified by the record. Not a rounding error.

Missing context

Literally true in isolation, misleading without the clause the headline dropped.

Unverified

Cannot confirm or deny with the sources available. Not the same as unknowable.

Hype index

A 1–10 score for the gap between evidence and story. High hype does not mean the technology is fake.

Low · 1–3

Modest gap between evidence and story, or the claim is mostly fair.

Elevated · 4–6

Real result, noticeable inflation in how it is sold.

High · 7–8

Large distance between what was measured and what the thumbnail promised.

Nuclear · 9–10

Category mistake territory. The facts and the mood are barely related.