D420 - sceKernelGetSystemSwVersion answers 13.09, which is not the 12.40 firmware

measured - 2026-08-31

Re-reading the module hardware reports (console-klog, console-report-klog, ps5-full-run) after "is there really nothing left" turned up a value this project had wrong. The D416 SwVersion work wrote 12.40.0.0 / 0x1240_0000 into get_system_sw_version, reasoning that 12.40 is "the firmware this machine models". The module dumps refute it: 130-layout/system-software-version shows the call writing the string 13.090.001 and the packed integer 0x1309_0001, identically across all three runs.

The trap was assuming one console has one version number. It has at least two, and they disagree:

• System software 12.40 - what obSCEne's sysinfo header carries (firmware|known|12.40), what the kern.version sysctl banner says (releases/12.40 Nov 27 2025), and what syscall 649 ([vendor_system_version]) answers from machine.firmware = 0x1240. All correct, all left alone.
• sceKernelGetSystemSwVersion = 13.09 - a different call reporting 0x1309_0001. Not the firmware, and deliberately not sourced from machine.firmware: wiring it there would make it answer 12.40, which is exactly the reconciliation D416 made and hardware refutes.

So the fix is not "change the firmware to 13.09" - the firmware (12.40) was right. It is to correct this one call's measured constant to 13.090.001 / 0x1309_0001, and to write the reasoning down loudly so the two numbers are not reconciled again. machine.rs already documented the distinction ("a conformance run read 13.090.001 from a different call ... not evidence about this one"); the kernel had simply not followed it. Byte layout was already right (D416's refinement - string at 8, int at 0x24, size word at 0 untouched); only the values were wrong.

Pinned with the_software_version_is_thirteen_oh_nine_and_the_size_word_is_untouched, which asserts 13.09 and that 12.40 is not what comes back - the regression that was made once. The wider lesson is the CLAUDE.md one about reporting more than the measurement supports: D416 took "12.40" from the project's own assumption, not from the dumped bytes, and the bytes said 13.09 the whole time. Kernel tests pass (72), clippy clean.