2026-09-01 - (/loop) Overnight crunch: putchar and std::random_device

The user asked for an overnight, oracle-free crunch of everything that does not need data. Triaged the remaining corpus walls first: they all trace to unimplemented functions answering placeholders, not logic bugs like the map one (D460), so the crunch is "implement the ones with a standard/citable contract, skip the hardware-gated ones, name the unnamed hashes". Aggregated the exact demand from the corpus run reports.

Implemented two, both from PPSA21564, both fully oracle-free:

• putchar - one byte to the output stream, returns it. Routed to stderr like puts/printf (D344). Published contract (ISO C / POSIX), known_by = published.
• _ZSt14_Random_devicev (std::_Random_device()) - the entropy primitive behind std::random_device. Implemented as a deterministic splitmix64 sequence for reproducibility; see D461 for why real entropy would be wrong here. known_by = assumed (the symbol-to-contract mapping is inferred).

cargo clippy/fmt clean on libc; libc tests pass (87); the knowledge audit passes (22) after two corrections worth noting for next time: the returns field is a typed enum (pointer/handle/count/status), not free text - omit it for a scalar-returning function as puts does; and an assumed entry must carry no cites (assumed means nothing confirms it), so the splitmix64/C++ references live in the source doc comment and the reasoning in assumptions.

Honest outcome: neither moved PPSA21564's wall - it still stops at the int 0x41 assert (image+0x11ccd) after exactly 500257 calls, unchanged. That number being identical across two runs is the useful signal: the deterministic random_device kept the run reproducible, where a real-entropy one would have made the count drift. So this was stub reduction (no more placeholder poison from these two), not a crack. The int 0x41 root is elsewhere - the flagged unimplemented sceKernelGetGPI is hardware-gated and the likelier cause.

Next oracle-free steps, in yield order: a naming pass on the 53 unnamed hashes (naming PS5Util::0xf948d02a4f9f5ace, 23413 calls, could unblock PPSA25872), then the remaining standard functions (vsprintf_s, and the advisory sceKernelSetVirtualRangeName / scePthreadSetaffinity whose effects are unobservable to the guest), then deeper wall analysis for logic bugs.