IOS LLE: checkpoint NAND conformance and diagnostic tooling

Preserve formatted NAND boot and Wii Remote pairing, add DMA/SRAM conformance tests and a Wii comparison probe. Keep P1 profiling tools and native update A/B evidence; reject the unhelpful targeted SRAM shortcut.

Tools/WiiConformanceProbe/README.md

@@ -0,0 +1,59 @@
+# Wii IOS conformance probe
+
+This is the first black-box comparison probe for the original IOS running on a
+real Wii and under this fork's Starlet LLE mode. Build **one** DOL and run that
+same file in both places. It records 64 rounds of IPC-backed resource-manager
+open/close calls in all four open modes, missing-device errors, root directory
+counts, installed-title counts, and an SD write/read/hash/cleanup round trip.
+The output contains 1,860 numbered observations. The probe requests no NAND
+writes, installs no IOS, and does not blindly scan MMIO registers. It creates and
+removes only `sd:/wii-conformance-data.bin` on the SD card.
+
+## Build
+
+Install devkitPPC and libogc/libfat using devkitPro, set `DEVKITPRO` and
+`DEVKITPPC` in the build shell, then run `make` in this directory. The output is
+`wii-conformance.dol`. On this Windows machine, the following command was used
+successfully from this directory in PowerShell:
+
+```powershell
+& 'C:\devkitPro\msys2\usr\bin\bash.exe' -c 'export DEVKITPRO=/c/devkitPro DEVKITPPC=/c/devkitPro/devkitPPC; export PATH=/c/devkitPro/devkitPPC/bin:/c/devkitPro/tools/bin:/usr/bin:/bin; make'
+```
+
+The ready-to-copy Homebrew Channel folder is `apps/wii-conformance/`, containing
+`boot.dol` and `meta.xml`. This build has not yet been run on a physical Wii.
+
+## Capture
+
+1. Use a Wii and a Dolphin LLE profile with the same IOS version and revision.
+   For exact directory/title counts, use the same NAND snapshot, starting state,
+   and loaded title. The SD cards need enough free space for a 4 KiB test file
+   and the CSV output.
+2. Copy `apps/wii-conformance/` to the `apps/` folder on each SD card and launch
+   it through the Homebrew Channel. Use the exact same `boot.dol` on both runs.
+   Dolphin LLE requires a configured writable virtual SD image.
+3. After the program returns, copy `sd:/wii-conformance.csv` from each SD card
+   to separate files such as `wii.csv` and `dolphin.csv`.
+4. Compare them with `python compare.py wii.csv dolphin.csv`. Exit code 0 means
+   all recorded observations agree, 1 reports differences, and 2 means the
+   captures are invalid or use different IOS versions.
+
+The CSV contains no keys or NAND contents. An allocated IOS file descriptor is
+normalized to success/failure because its numeric value depends on unrelated
+open handles. Call return codes and output counts remain exact. Time is not
+compared in this first suite because scheduling and output I/O introduce noise.
+`captures/` is kept local and ignored by Git until each capture's console or
+emulator provenance, IOS revision, probe binary hash, and initial NAND state
+are documented. A filename alone is not evidence that a capture came from a
+physical Wii.
+
+## Interpretation and next suites
+
+A mismatch identifies a reproducible behavior to investigate; it does not by
+itself locate a bad register. Trace the first failing request inside Dolphin's
+Wii IPC, Starlet memory bus, and the relevant device. The next test suites
+should use explicit stimulus/response cases for IPC flag acknowledgements,
+shared-memory cache visibility, DMA, IRQ ordering, and individual devices.
+PPC homebrew cannot inspect every Starlet register while Nintendo IOS is
+running. A separate Starlet-side probe under MINI/BootMii can test raw
+hardware registers, but its results must be labeled as MINI-mode tests.