D097 - The wrap default was justified by a population counted in the wrong place, and the default is right anyway

Status: decided. 2026-09-09.

Its subject no longer exists (D100). wrap is retired, and the pipeline that took its --privilege and --sdk requires --target, so there is no default magic left for this reasoning to justify. What follows stands as the record of why the default was right while it existed.

selfish wrap defaults to the previous generation's magic. The reason given for that was a count of containers that are all of homebrew lineage, and a hardware citation that turns out to name a fake container. The default does not change; its justification does.

Two places said the same thing:

What is wrong with both

The data table's citation is a fake container. PPSA02664 measures ptype 0x1 in the 32-container census of sweep 20260909-175052 - the fake marker. So the row confirms that a fake eboot carries the previous generation's magic, which is unsurprising: the writer this table was derived from emits exactly that. It is the same circularity D092 was reversed for, sitting in a row rather than in a verdict.

The same probe's census points the other way. Forty installed eboot containers: 38 with the current magic, 2 with the previous one. Whatever the population says, it does not say what the row claimed.

And the CLI's thirty-three were counted inside packages. A package is what this collection and the homebrew store build; containers found inside them are of that lineage. "Including a working homebrew store" was offered as strengthening the claim and is the tell that it was the wrong population - it is more homebrew, not less.

Why the default stays 4

Because the evidence that actually matters was never the population. A package this repository builds with the previous generation's magic installs, mounts, loads and executes on current hardware (worklog 040). That is a measurement of the thing the default is for - whether the file is accepted - and it is untouched by any of the above.

So the default is unchanged and better supported than it was, because "it runs" is a stronger reason than a count taken from the wrong shelf. Both notes now say that instead.

What is not being decided

Whether a genuine current-generation eboot carries the current magic. The census suggests it, 38 to 2, but the census does not report magic per container keyed by ptype, so it cannot separate genuine from fake - the same confound recorded against the five header rows. Filed as a request rather than concluded here. Principle 5: current_magic and previous_magic keep their values, and the claim about which file carries which is a row of prose that has been marked under review rather than replaced by a better guess.

The general point, which is the fourth of its kind today

The citation was obscene 048-selfaudit, PPSA02664 - a real probe, a real sweep, a real container id. Everything about it looked like evidence, and the one thing nobody checked was what that container is. Reading a name as a fact, again: a candidate label, a verdict headline, a field name, a doc comment, and now a title id in a citation.

Settled, 2026-09-09: refuted, and the default still does not move

REQ-20260909T2020Z-8a17 asked for the magic keyed by ptype. The 2x2 across all 32 containers (sweep 20260909-184538):

current magic previous magic
ptype 0x0 (genuine) 23 0
ptype 0x1 (fake) 7 2

The withdrawn claim is refuted rather than merely unsupported. Every genuine container carries the current magic; none carries the previous one. And its evidence, PPSA02664, is one of the two containers on the box carrying the previous magic - both of which are ptype 0x1.

Here the confound really is broken, which is worth stating because the five header rows are still stuck behind theirs. The bottom row is the control group: the fake population spans both values, so carrying the current magic is not something the console imposes on whatever is installed. Genuine 23/0 against fake 7/2 is a correlation that survives having a comparison.

And the default stays 4, which is the same discipline as the entry above

The entry above removed a population claim from the justification. Accepting a different population claim now would be that error in reverse.

Nothing measured says whether a file built with the current magic would be accepted. What is measured is that a package built with the previous one installs, mounts, loads and executes (worklog 040). Acceptance is the property the default exists to secure; what other people's files contain is not, in either direction.

That 7 fakes carry one value and 2 carry the other says the console tolerates both from a fake container - which is consistent with the default being a free choice and is not evidence that it is. The thing that would move it is a file built here with the current magic, run.

Settled again, 2026-09-09: both magics are accepted, and the default becomes a route decision

REQ-20260909T2125Z-3f9d asked whether obSCEne had already run a container built here with the current magic. They had, twice over:

So the 2x2 in the section above already contained this repository's own output as one of its seven fake/current-magic entries, and nobody knew.

What that changes, and what it does not

Both values are accepted. The reason the default stayed 4 up to now - that it was the only value shown to be accepted - is gone.

The default stays 4 anyway, on a different reason: it is a route decision. The two proofs are on different delivery routes and neither is on the other's:

route magic evidence
package previous installs, mounts, loads, executes (worklog 040)
native title directory current installs, launches, 292 checks (sweep 20260909-184538)

wrap's default serves the package path. The native path passes --generation 5 explicitly and never consults it. Moving the default would put the package path onto a value nothing has accepted on that path, and D082 already records that the two are different delivery routes rather than two spellings of one.

The line that survives all three revisions

The default has never moved on a population, in either direction. Not when the population was claimed to say previous - that claim was withdrawn, and its evidence turned out to be a fake container. Not when the population turned out to say current, 23 of 23 with a control group, which is a far better measurement than the one withdrawn.

A default turns on what a loader accepts, and the answer differs by route. Three revisions of this entry in one day, and that sentence is the only part that did not need rewriting.