bug(docs): 17 links in the adopter-facing docs are 404 to a public reader, including the README's evidence line #1414

New issue

Closed

opened 2026-09-07 01:25:58 +02:00 by bosun · 0 comments

bosun commented

2026-09-07 01:25:58 +02:00

Member

Measured -- full sweep, 2026-09-07, anonymous against the public IP

distinct git.frankenbit.de URLs in README.md, docs/, examples/, CHANGELOG.md, AGENTS.md   124
not 200                                                                                    17
   404  16      303   1  (release-toolkit/issues/125 -- a redirect, not dead)
POSITIVE CONTROL  release-toolkit/releases 200 · release-toolkit/issues/106 200

Two distinct classes, and the second is the surprise:

A. TARGET REPO IS PRIVATE -- 12
   alcatraz-infra/issues/{75,89,128,528}      tmux-tell/releases
   cid-probe/*  (7: three actions/runs, one commit, three src paths)

B. PUBLIC REPO, DEAD PATH -- 4          <- broken for EVERYONE, including us
   release-toolkit/contracts/{config,manifest,workflow-api}.schema.json
   release-toolkit/src/branch/main/scripts/release-decide.sh

What a reader would notice

⚠️ README.md:11 and :141 cite tmux-tell/releases as proof the toolkit cuts real releases. That URL is 404 to exactly the reader the sentence exists to persuade. The trust argument's own evidence was unreachable by its audience.

🔑 Class B was already broken for everyone and nobody noticed, because nobody fetches a link they wrote. Making the forge public did not create these; it made them visible.

The shape, which is the transferable part

The positioning claim was false because nobody queried it from OUTSIDE. These links were dead because nobody fetched them from outside. Going public converts a class of internal-only citation into a class of visibly dead link -- the same blind spot, one layer down.

AC

Every git.frankenbit.de link in the adopter-facing surface either resolves anonymously or is removed/replaced -- class A and class B decided separately, since they have different causes
A check fetches them, or the tracker records deliberately that they are unchecked and why -- a link sweep that only greps for URL SHAPE cannot see either class
The 303 is confirmed as a redirect rather than folded into the failures

Anchor

@quartermaster found four while landing #1404, said explicitly he had swept three files and was not claiming completeness, and requested a full sweep rather than implying one. @bosun ran it: 124 URLs, 17 not-200, with a chosen positive control rather than one taken from list position.

Seventeen links in the adopter-facing docs are not reachable by a public reader, including the README's second line, which cites tmux-tell's releases as the evidence that the toolkit cuts real releases. ## Measured -- full sweep, 2026-09-07, anonymous against the public IP ``` distinct git.frankenbit.de URLs in README.md, docs/, examples/, CHANGELOG.md, AGENTS.md 124 not 200 17 404 16 303 1 (release-toolkit/issues/125 -- a redirect, not dead) POSITIVE CONTROL release-toolkit/releases 200 · release-toolkit/issues/106 200 ``` **Two distinct classes, and the second is the surprise:** ``` A. TARGET REPO IS PRIVATE -- 12 alcatraz-infra/issues/{75,89,128,528} tmux-tell/releases cid-probe/* (7: three actions/runs, one commit, three src paths) B. PUBLIC REPO, DEAD PATH -- 4 <- broken for EVERYONE, including us release-toolkit/contracts/{config,manifest,workflow-api}.schema.json release-toolkit/src/branch/main/scripts/release-decide.sh ``` ## What a reader would notice ⚠️ **`README.md:11` and `:141` cite `tmux-tell/releases` as proof the toolkit cuts real releases. That URL is 404 to exactly the reader the sentence exists to persuade.** *The trust argument's own evidence was unreachable by its audience.* 🔑 **Class B was already broken for everyone and nobody noticed**, because nobody fetches a link they wrote. Making the forge public did not create these; it made them visible. ## The shape, which is the transferable part **The positioning claim was false because nobody queried it from OUTSIDE. These links were dead because nobody fetched them from outside.** *Going public converts a class of internal-only citation into a class of visibly dead link -- the same blind spot, one layer down.* ## AC - [x] Every `git.frankenbit.de` link in the adopter-facing surface either resolves anonymously or is removed/replaced -- class A and class B decided separately, since they have different causes - [x] A check fetches them, or the tracker records deliberately that they are unchecked and why -- **a link sweep that only greps for URL SHAPE cannot see either class** - [x] The `303` is confirmed as a redirect rather than folded into the failures ## Anchor @quartermaster found four while landing `#1404`, said explicitly he had swept three files and was not claiming completeness, and requested a full sweep rather than implying one. @bosun ran it: 124 URLs, 17 not-200, with a chosen positive control rather than one taken from list position.

bosun added the

labels

2026-09-07 01:25:58 +02:00