12. N-way consensus as a substitute oracle
Nothing carries hardware, and on current trajectory
nothing will until the hardware is available. That caps what the suite
can claim: when obSCEne and an emulator disagree, the report cannot say
which is wrong, and fail [assumed] is a verdict the
recipient can decline without argument.
There is an oracle available that is not the hardware:
agreement between independent implementations.
obscene-tool consensus, taking N reports and printing only
where they disagree, would change the message from "you failed check X"
to "you are the only one of four that fails X". That is actionable
without authority and costs a subcommand.
It also gives the provenance ladder a rung that can be climbed
without hardware: assumed → consensus →
hardware.
Two things it must not pretend. These emulators are not independent of each other - they read each other's source, and fpPS4's NID table is a superset of ps4libdoc's. This project made exactly that mistake once (D064), counting one source twice. So consensus must record which implementations agreed, so correlation stays visible rather than being laundered into a number.
The prerequisite turned out to be softer than this
said. It claimed consensus needs a second emulator.
The host build is a real implementation of the POSIX and C library
surface, so disagreement against it is disagreement with something known
to work - which is the closest thing to an oracle available without the
hardware. consensus runs on host and shadPS4 today.
And Kyty is closed as a candidate, with evidence
(D080). Its sceKernelWrite refuses standard output by
design, and the puts fallback is registered under a library
name this module does not import from. Getting a report out of it would
mean importing the C library under a loader-specific name - an
accommodation in the import table itself, which is worse than the three
that already exist. It remains the only source of a named
unresolved-import list, which is what it is uniquely good at.
Overtaken. This said craziiEmu was "the remaining
candidate for a third report" - there are now five reports, from
shadPS4, PS5PCEM, fpPS4, Kyty and prosper, and craziiEmu is not among
them. The scarce thing is no longer a third report; it is a third
current-generation one: PS5PCEM and prosper both complete, and
two implementations produce only SPLITs and never an
OUTLIER, so no majority can form and neither can be called
wrong. An oracle needs three. (D197)