The filesystem is cleared, and there are four walls

Asked whether the data the guest loads is actually correct. Verifying content would double every read; verifying completeness costs a counter and catches the failure that matters - a truncated asset that faults later inside the guest's own parser (D175).

One distinction makes the count useful: reading to the end of a file is a short read by definition, so only reads cut short before the end are counted.

files    10 reads, 11328 KiB, none cut short

Every byte the guest asked for arrived. The filesystem is exonerated - worth as much as a bug would have been, because it removes a layer from suspicion.

Then ran the whole corpus, which said something the single title never could: the wall was never singular. Four titles, four walls, in four different places - plus two that do not parse at all, being previous-generation containers. image+0x43c4 had started to feel like the blocker, and running one title repeatedly is exactly how that happens.

The better lead now is PPSA03416: an illegal instruction three frames from the entry point with our error code sitting in rax, executing twenty-two megabytes into the image. That reads as the guest calling through a pointer a stub handed it - much closer to the surface than Earthion's wall.

And the previous-generation parse failure is the only item that changes the denominator: four runnable titles out of six, not one wall out of one.