ac72e9ac4e
EpsonTest predates the adapter layer: PrintCommandsToEpsonPrinterAsync takes an EpsonPrinter and drives it directly, so PrinterWrapperFactory, the model adapters and the 0x20 dispatch were never exercised. Against a POS895 that path feeds with ESC d, which the firmware ignores, so every cut slices through the last printed line, and it never sends a logo at all - the only logo branch is the HTML one, which uses GS ( L, the command that prints nothing on this hardware. PrintCommandsViaAdapterAsync instead does what EpsonPrintService does for a real job: query GS I 1, let PrinterWrapperFactory choose the adapter from the reported byte, then print the logo and commands through it and cut. It logs the reported id and the chosen adapter, so a wrong or missing model byte is visible rather than silent. Menu gains a third target and the non-interactive form accepts "pos895", so `dotnet run -- final pos895` prints a Final receipt with its logo through the production path. That covers the one part of the POS895 work nothing had executed end to end: the ESC * bit-image logo. Note EpsonTest must be run from its own directory; the logo paths are relative and `dotnet run --project` from the repo root cannot resolve them. That predates this change. Claude-Session: https://claude.ai/code/session_01E6N3P4nSZLtp4BC2LXVref
