UNREPOST

Time & language

Times

UTC is the default. Local uses your device's time zone. Dates without a known time stay the same.

Language

Translates this page in place with Google Translate and keeps it on as you browse. Machine translation can be wrong; the English text is the reviewed record.

All settings

Public documentation

Methodology

How sources are found, checked, labelled and published, and how we keep a critical stance from overriding accuracy.

UNREPOST takes a critical editorial stance on public conduct and product decisions. It also holds itself to the rules below, and those rules win whenever they conflict with the stance.

1. Finding sources

The research pass for release 2026.10.28-r1 started from a fixed list of starting points: the court filings, the decision, plaintiffs’ counsel’s case page, and named news and analyst reports. For each one we:

  1. opened it, or recorded that we could not;
  2. read the relevant section, such as PDF pages 104–107 of the complaint and pages 3, 45–50 and 77–79 of the opinion;
  3. followed embedded links to originals where possible, such as analyst posts, X statuses and market pages;
  4. recorded publisher, author, displayed date, page or figure, access result and limitations.

Unrelated accusations are not added just to make the archive bigger.

2. Separating where from what

Every source gets a provenance label for where it came from. Every receipt gets an editorial status for what the material establishes. A screenshot inside a complaint is a primary document, a court filing. What it establishes is that it was published in that filing, not that the underlying chat is authentic or complete. See labels.

3. Attribution and adjacency

  • Serious claims name who makes them in the same sentence: “Lookonchain, as reported by crypto.news, estimates…”, never “insiders made…”.
  • A source sits next to every material claim.
  • Wherever the complaint’s legal theory is summarised, the court’s decision is linked beside it. The build fails if it is not.

4. What does not follow from what

These inferences are not made:

  • A repost and its removal do not by themselves show coordination, intent, wallet ownership or profit.
  • A claim that an account was hacked stays unconfirmed without an attributable statement or independent evidence.
  • Market capitalisation is not cash stolen, deposited or lost.
  • Similar timing and wallet transfers can support analysis. They do not establish a person’s identity or state of mind.
  • A pending civil allegation is not a criminal conviction.
  • Not finding a response does not mean none exists. The search date and scope are always stated.

5. Numbers

When observers disagree, such as ~$14M against ~$15M or >95% against >98%, each figure is listed with its observer and time. Figures are never averaged into a new number. Figures about different token mints are never combined. No loss total is computed across dossiers.

6. Quotes

Quotes are short, live in one canonical record each, and keep their context. Figure quotes were transcribed by eye from the images embedded in the filing. Speaker labels are reproduced as shown, such as “Alon” or “out”, and the mapping from label to person is attributed to the complaint.

7. Images

  • Court figures are the images embedded in the filed complaint, extracted from the PDF. They are unaltered except for a provenance strip and one marked redaction of abusive third-party text.
  • Post screenshots show the original x.com posts captured on 28 October 2026 (UTC). Each preserves the visible post date/time, our capture time and a provenance strip. Separate carousel views are labelled before being joined vertically. Earlier capture-service and embed source records remain in the history.
  • The default post view covers photos of people, scannable QR codes and personal links with visible, labelled boxes. Settings offers the original post screenshots without these redactions. Both views retain the reviewed crop and provenance strip, and each lists its own edits. Court and market-image redactions remain as labelled.
  • Deleted posts that nobody captured are not pictured. Nothing is reconstructed or generated. We do not get past bot checks or sign-in walls to take a screenshot.

8. Dates

Event, publication and retrieval dates are stored separately, with explicit precision. See dates and sources.

9. Build-time checks

Before publication, scripts/validate-content.ts checks that:

  • every reference resolves;
  • every quote comes from a source we opened;
  • page links point into the registered PDF at the labelled page;
  • precision matches the stored form;
  • mints are valid, distinct, and never the project’s own;
  • conclusory words such as “stolen”, “proven” or “convicted” appear only in negated sentences;
  • every receipt has a revision history;
  • every image file exists at its recorded size, comes from a source we opened, and is cited by each receipt that shows it.

A failing check stops the build.

10. Publishing and corrections

Reviewed content lives in the repository. A new commit builds a new public release with a versioned manifest. Correction submissions go to a private queue and never publish automatically. Changes appear as dated revisions, and superseded wording is kept. See corrections.

11. Tone

Satire is welcome in the meme lab, and every image it produces is labelled as satire. Factual pages do not joke. No content targets anyone’s religion, nationality, ethnicity, family or unrelated personal characteristics, and no private contact details are published.