2026-08-31 (later) - C succeeded: the full current obSCEne runs as the payload

The payload shape (obscene.elf, the 9.3MB artifact the corpus asset actually names, and the exact bytes hardware's payload run streamed) runs the current obSCEne in orbistoun: meta 1|38|543, 189 imports, 29/38 sections, 158 unique tests captured through the now-implemented sceKernelDebugOutText (+ fd1). No handoff needed - it carries a real import table, so the plain entry runs it. This is the valid-diff unblock the stale corpus (Aug-25, 27 sections) had been hiding.

And it settles the earlier "missing tests" question outright: 130-layout/memory-type now runs and is pass 0x3, query-size-ladder partial, short-buffer-overrun pass 0x4 - byte-for-byte the hardware verdicts. They were never an orbistoun gap, only the stale binary. B (the eboot's PT_SCE_DYNLIBDATA vendor-segment fault) is therefore optional: the payload, not the eboot, is the corpus artifact.

New blocker, narrower: the run crashes ~29% in, in the video section. sceVideoOutGetFlipStatus, sceVideoOutSubmitFlip and sceVideoOutSetFlipRate are declared but not in video::implementations(), so a flip is submitted, its completion polled, and the placeholder read into a fault. That is a video-flip model to build (a new subsystem piece), flagged rather than assumed. Capture note: obSCEne emits each record to both DebugOutText and fd1 and orbistoun forwards both to stderr, so the raw stream double-counts - dedupe on read.