Catalogue · Imprint · Superseded edition
This edition is out of force. It governed the catalogue from 2026-06-02 to 2026-08-14, when the second edition replaced it. It is kept in print because a state set under a rule cannot be audited against a different rule, and because 107 entries in this catalogue moved when it was replaced.
The rubric, first edition
First edition. In force from 2026-06-02 until 2026-08-14. It rated repositories only: an entry with no repository carried no state at all, because the rules had no field to read. Four states, one threshold pair, and no link liveness sweep.
The thresholds it applied
| State | Set when | Note |
|---|---|---|
| ACTIVE | pushed within 180 days of the field check | Half a year. Generous, and it is the clause the second edition tightened. |
| SEMI | pushed 180–1095 days before the field check | One bucket covering everything from a slow quarter to a three-year silence. |
| DEAD | pushed more than 1095 days before the field check | Three years. |
| ARCHIVED | GitHub archived flag is true | A maintainer's own statement, and it outranks every date rule. |
| — (unstated) | no repository | Test pages, papers, vendor docs and communities carried no state under this edition. |
What the second edition changed
| Change | First edition | Second edition |
|---|---|---|
| ACTIVE threshold | 180 days | 90 days |
| States available | 4 | 6 — NRND and WITHDRAWN added |
| Non-repository entries | no state | state from the link check |
| Source fields read | 1 endpoint | 3 endpoints |
Applied to the same field data, those four changes moved 17 entries from one state to another and gave 90 entries a state for the first time — 107 of 153 in all. Because both editions are applied to one field check rather than to two different sweeps, the difference between them is a property of the rules alone. The catalogue's AS PUBLISHED / CORRECTED toggle shows exactly that, and the errata leaf names every entry it moved.
What a superseded edition is still good for
Three things, all of them audit rather than history. It lets you take any state this catalogue printed before 2026-08-14 and check it against the rules that were actually in force when it was printed. It lets you see which of the entries that moved were moved by a threshold and which by a clause that did not previously exist. And it makes the register's own claim falsifiable in the strongest available sense: apply these thresholds to the columns in the published data and you must reproduce the state in the state_first_edition column, row for row.
That column exists in the payload for exactly this reason. A register that publishes only its current verdict is asking to be trusted; one that publishes the verdict, the superseded verdict, the rule behind each and the fields both read is asking to be checked instead. Reading the NRND group is the clearest illustration: most of those entries carried no such mark under the first edition, because the first edition had no such mark to give.
Questions about superseded editions
Answered from the two rule sets printed above and from the errata they produced.
Why keep a superseded set of rules in print?
Because a state set under one rule cannot be audited against a different rule. Every entry whose state changed carries both derivations on its own page, and the errata leaf prints the superseded value struck beside the replacing one. Delete the old edition and every one of those records becomes an assertion nobody can check.
Did the tools change, or did the rules change?
The rules. Both editions are applied to the same field check rather than to two different sweeps, which is what makes the difference between them a property of the rules alone. Nothing in the world moved between the two columns above; only what this register was willing to say about it did.
What did the first edition get wrong?
Two things, and both are structural rather than arithmetic. It called anything pushed within 180 days ACTIVE, which is longer than the cycle detection changes on. And it could only rate repositories, so every test page, paper, vendor document and community in the catalogue carried no state at all — which meant the entries most likely to go quietly dead were the ones it said nothing about. The edition in force fixes both.
How does a new edition get numbered?
A new edition when a threshold, a source field or an endpoint changes — when the rule changes. A new printing when only the data is re-swept — when the world changes. The folio at the foot of every page carries both ordinals, and they answer two different questions. The field log records every sweep.
Publishing a method, versioning it, and keeping the superseded version in print is the one axis nobody in this field competes on. Donut Browser is the tool the people who keep this register work on, and it is held to the same standard here as everything else in the catalogue: an open-source anti-detect browser under AGPL-3.0, with the desktop application and the sync server both in the repository, engine-level fingerprint spoofing through the Wayfern Chromium fork rather than injected JavaScript, zero telemetry, and a pre-launch gate that refuses to start a profile when the proxy's exit geography contradicts the profile's fingerprint. Its row moves when its fields move, under whichever edition is in force.
curl -fsSL https://donutbrowser.com/install.sh | sh
macOS · Windows · Linux · Nix. Desktop only. Source, AGPL-3.0 · Its entry · The keepers' disclosure