2026-09-09 - The magic, settled by the row nobody asked for

Worklog 057 marked the current_magic / previous_magic rows under review and filed REQ-20260909T2020Z-8a17 for the magic keyed by ptype. It came back, and it is the most decisive measurement of the day:

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

Every genuine container carries the current magic; none carries the previous one. And the claim withdrawn earlier that afternoon - that a current-generation title's eboot carries the previous magic, marked hardware-confirmed - is now refuted rather than merely unsupported: its evidence was PPSA02664, which is one of the two previous-magic containers, and both are fake.

What made it settle, which is the part worth carrying

The bottom row. It was not asked for - the request wanted the magic beside ptype, and the 2x2 fell out of that - and it is the whole reason this question is closed while the five header rows are not.

Genuine 23/0 on its own proves very little. "Every container on a current-generation console carries the current magic" would explain it without reference to authenticity at all, and this repository has spent the day being caught by exactly that class of alternative. The fake row is what removes it: 7 carry one value and 2 carry the other, so the console does not impose the current magic on whatever is installed. The correlation survives having something to compare against.

That is precisely what the five header rows lack. Every ptype 0x1 container there is PPSA-shaped and every ptype 0x0 one is NPXS-shaped, with no case on the other diagonal - so there is no control and the confound stands. The difference between a settled row and a stuck one, on this console, is whether the sample happened to contain a counterexample class.

The default did not move, and that is the same discipline twice

D097 removed a population claim from the wrap default's justification in the morning. Accepting a different population claim in the evening - "genuine containers carry the current magic, so write that" - would have been the identical error reversed, and it was tempting precisely because the new measurement is so much better than the old one.

The default turns on acceptance, not on population. What is measured is that a package built with the previous magic installs, mounts, loads and executes (worklog 040). Nothing measures whether a file built here with the current magic is accepted, so the default stays where the evidence is.

Filed as REQ-20260909T2125Z-3f9d - and that request is mostly a retraction of a guess. obSCEne's Makefile defaults EBOOT_GEN to 5 on the Prospero path and passes it to mkself --generation, and their eboot leg runs, which looks like the measurement already exists. It was asked rather than concluded: a Makefile default, a fix described in a comment, and a log line reading generation|known|5 (agc) that is the probe detecting the console rather than reporting its own container, is the same three-input inference that has cost five corrections today.

One protocol note

obSCEne re-filed under REQ-20260909T1715Z-5b91, an id already resolved in this inbox, so that id now appears twice - once resolved, once open. The resolved block was not touched and the new content was serviced as a new ask, but commit messages here cite request ids, and an id that resolves to two blocks cannot be cited. Flagged back to them.