455. The biggest number on the board

2026-09-08 - directed, continuing 454

Bus idle a fourth pass. So: worklist, which is the tool that exists for "what next" and which this session had not run.

One import is 87.6% of every call this project has ever recorded

      CALLS  SHARE  MODULES  IMPORT
   19689012  87.6%        1  PS5Util::0xf948d02a4f9f5ace
     595200   2.6%        6  libkernel::sceKernelDlsym

Nineteen and a half million calls, one guest, no name - PPSA25872, at 2% standing, the worst in the corpus. The report calls the shape correctly: a guest that keeps asking the same question has not accepted the answer.

PS5Util is not a platform library. It is /app0/Media/Modules/PS5Util.prx, a file the game ships, and orbistoun places, relocates and starts it. So there is nothing to implement: the code the guest wants is code the guest brought, mapped in this process.

Cross-checking exports against imports:

module	eboot imports	module exports	matched
PS5Util.prx	2	7	2
Il2cppUserAssemblies.prx	4	279	4

Six for six. Every import the executable makes from a module the title ships is exported by that module - 0xf948d02a4f9f5ace is PS5Util + 0x2a0 - and all six are answered with the Unimplemented placeholder instead (D630).

Left as a diagnosis, and why

Resolving an import to a placed module's export is loader work, and this session's brief confines the work to user-space library and ABI-stub implementations. The diagnosis is complete enough to act on without more investigation - module present, placed, started; NID in its export table at a known offset; no measurement, name or capability missing.

And there is a real question under it that stops "just link them" from being the obvious answer: the same title ships libc.prx, and 186 of the executable's imports come from it. For those, orbistoun's own implementation is almost certainly better than the game's copy, and D005's whole architecture is that an import is intercepted. So the rule has to separate a library this project implements from a module only the game could have written. PS5Util and Il2cppUserAssemblies are plainly the second kind; libc.prx is plainly the first.

Surprises

• The corpus's biggest number was never looked at. 87.6% of all recorded calls, sitting at the top of a ranked list that names itself "what to implement next", and it turned out to need no implementation at all.
• The spin's arguments say it is a list operation. arg5 points at a structure whose +0x10 field holds its own address - an empty intrusive list, next = self - and arg0 is a host-side address outside every region this run gave the guest. Recorded rather than read into.

Next

• The six title-module imports, when loader work is in scope. This is the largest single wall measured anywhere in the corpus.
• The unapplied relocation behind obs_sink_open (D628) - the same class of thing, one layer down.
• The declined-syscall casualties, waiting on REQ-20260908T1620Z-4c1e.
• sceAgcCreateShader, waiting on REQ-20260908T1621Z-7a5d.