I have reached the limit of what I can do. To me, this looks like a problem caused by concurrent access (likely writes) to the SQLite database, where the WAL-journal method is supposed to handle mutual exclusion.
Since both you and Claude insist on an eMMC fault—and given the doubts surrounding the MMC “Life Time” and “EOL” values which are available—only @sam_nazarko can take this further and perhaps suggest a reliable way to test the eMMC.
The SQLite version being used is far behind the current release; versions 3.44.6, 3.50.7, and 3.51.3 (2026-03-13) include a fix for a rare WAL bug that causes database corruption. However, it is unclear whether the symptoms observed here are actually caused by that bug. I am handing the baton over to @sam_nazarko.