410. The tail, and what was decided not to do

2026-09-04 - directed, while the Agc probe set is out for hardware

What was done

Eighteen non-Agc functions were called and unimplemented, twenty-two calls between them, almost all once each. Ranked by call count there was nothing there. Ranked by what the placeholder does when the guest reads it, two were dangerous:

• setjmp answered 0x7fff_0001, which is non-zero - and a non-zero setjmp tells a program it arrived via longjmp, sending it down an error-recovery path it was never on.
• setlocale answers a char *, so the placeholder was a wild pointer the guest dereferences. The D125 shape, again.

Implemented six: setjmp, setlocale, clock, _sigprocmask, sceKernelUuidCreate, sceVideoOutConfigureOutput. Down to 20 unimplemented, 22 stubbed calls from 35 and 914 at the start of the day.

The larger half was deciding not to

Nine of the eighteen turned out to be recorded decisions rather than gaps, and reading the record before writing code is what found that:

• sceUserServiceGetLoginUserIdList - D346 already decided it, in the declaration itself: it writes a list whose layout is unmeasured. Implementing would have reversed a decision on no new evidence. Fifth time today that reading the existing record changed what I did.
• sceCommonDialogInitialize, sceMouseInit, sceAppContentInitialize - reporting that a subsystem initialised when it does not exist is the exact stub principle 3 forbids. A guest told the mouse is ready calls sceMouseOpen next and gets a placeholder.
• sceVideoOutSetBufferAttribute2, sceVideoOutGetOutputStatus - fillers, not setters; both promise to write a structure no lawful source here lays out. ConfigureOutput is implemented because it promises nothing about the caller's memory, which is the line between them.
• scePthreadAttrGet / AttrGetstackaddr - D561's unexplained argument, still unexplained.
• PS5Util::0xa96b2b178383025c - no name, nothing to implement.

Everything still unimplemented is now either out for hardware (eleven Agc and Ampr) or one of those.

Surprises

A gate caught me twice in one session. every_implemented_function_is_written_down failed after the pthread batch and again after this one - six knowledge entries missing each time. It is the second-most useful thing in the tree after the guest itself, and both times it was faster than noticing on my own.

setjmp with no longjmp. The binary imports setjmp and never imports longjmp, which is what makes answering zero safe today and is written into the entry as the condition to re-check. Implementing longjmp without first implementing the save would be the dangerous change.

What is not claimed

Each of the six is better than the placeholder it replaced, which is a lower bar than correct. setlocale answers the C locale because orbistoun implements no other, not because the console does. clock's unit is a coin toss between POSIX's 1,000,000 and FreeBSD's historical 128 - a factor of 7,800, recorded as the first thing to change if a title's timing looks wrong.