D637 - An answer needs somebody who asked

Status: measured Date: 2026-09-09

A finding I added, at the head of a list it had no business being on

D625 introduced Gap::Captured and printed it ahead of the ranked six, on the argument that a finding which exists because somebody typed a variable is not competing with findings the tool volunteered. That argument is right and the implementation did not honour it.

An ordinary run of the conformance eboot - no diagnostics, nothing asked for - now opened with:

what to do about it
  ! libc::acos was asked about, and here is what it was passed
      12 captured argument value(s), listed below
  ! libc::asin was asked about, and here is what it was passed

Nobody asked about either. The real finding - libScePad::scePadReadState called 180 times with nothing implementing it - was pushed down the page by two functions that work.

Why those two, and why it was invisible

captured filtered on having a dump, and dumps are taken for every import nothing implements as well as for anything forced. So the question is which imports have a dump and no unimplemented finding to claim it, and the answer is a category nobody had thought about:

The dump condition tests handler.is_none() - the integer handler. A function answering in xmm0 is bound to the float table (D268), so its integer handler is None, so it is dumped by default - and it is implemented, so unimplemented produces no finding for it. Exactly the maths library, exactly acos and asin.

Two tables that are disjoint by construction, and one condition that only knows about one of them.

The forced list is now published, not just counted

arm_dumps already computed which slots a run named; it printed the count and threw the labels away. They are recorded on the trace now, and captured fires only for them.

ORBISTOUN_DUMP=sceKernelWrite still produces exactly one captured finding; an ordinary run produces none, and scePadReadState is back at the top where it belongs.

Both directions, tested

A version that emitted nothing at all would satisfy "no noise in an ordinary run" perfectly - and emitting nothing is precisely what this finding did for its entire first day (D625). So the test asserts both: a dump taken by the default rule produces no Captured finding, and an import somebody named still produces one.

The thing worth keeping

This is mine, made today, in the same session as eight records about instruments that could only say one of the two things they appeared to say. It was found by running a guest for an unrelated reason and reading the top of its report.

The lesson is not "be more careful". It is that the report is the instrument that checks the other instruments, and the only way to use it is to keep reading ordinary runs of guests nobody is currently investigating.