2026-09-09 - A title id in a citation, read as a fact

Quiet tick, nothing outstanding, so: the suffix file had carried a false claim about a sibling repository for however long, and the check that found it was reading twenty lines of theirs. Are there others? Bounded version - grep the data/ files, which are the source of truth, for claims about other projects.

Nine hits. Seven are provenance citations or this session's own measurements. Two are the generation-magic rows, and they are wrong in a way that took ten minutes to establish and would never have been noticed by reading them.

The claim

data/self-format.tsv, current_magic: "a current-generation title's eboot carries previous_magic (4F 15 3D 1D), measured on hardware (obscene 048-selfaudit, PPSA02664: gen4 eboot=1, gen5=0)". And previous_magic: "what a current-generation title's EBOOT actually carries (hardware-confirmed)".

PPSA02664 measures ptype 0x1 in today's census. It is a fake container. So the citation establishes that a fake eboot carries the previous generation's magic - which is what the OpenOrbis-lineage writer this table came from emits. The same circularity D092 was reversed for, this time living in a row rather than in a verdict.

And the same probe's census counts, across forty installed eboots, 38 with the current magic and 2 with the previous one. The population says the opposite of what the row claims.

What rested on it

selfish wrap --generation defaults to 4, and its help said: "4 is the default, and that is a measurement rather than a habit. Every container found inside real packages for the current console carries the previous generation's magic - thirty-three of them, including a working homebrew store."

Thirty-three containers found inside packages. Packages are what this collection and the homebrew store build, so containers inside them are of that lineage. "Including a working homebrew store" was written to strengthen the claim and is the tell that the population was wrong - it is more homebrew, not less.

The default does not move

Because the evidence that matters was never the count. A package this repository builds with the previous generation's magic installs, mounts, loads and executes on current hardware (worklog 040). That measures the thing the default is for - whether the file is accepted - and nothing above touches it.

So both notes now rest on that instead, and the default is better supported than it was. "It runs" beats a census taken from the wrong shelf. What is not decided is whether a genuine current-generation eboot carries the current magic: the census does not report magic per container keyed by ptype, so it cannot separate genuine from fake - the same confound already recorded against the five header rows. That goes out as a request. (D097)

Fifth

A candidate label, a verdict headline, a field name, a doc comment, and now a title id inside a citation. obscene 048-selfaudit, PPSA02664 looks exactly like evidence: real probe, real sweep, real container. The one thing unchecked was what that container is - and this repository had spent the afternoon building a tool that answers precisely that question and printing it above the row count so nobody could miss it.

It was missed anyway, in a file the tool does not read.