PR15030 native keyboard proof, reviewed head10a559eb78969c6b4338baa550a8383506ffb0a2
Production22961e3d58ea6ed42164d6ece84358b042dadefb; frozen tests c1c3138b9ba302ca8624066c64eaf174ddb97d85.
Actual Robolectric4.16.1/API36 on task-local JDK21/SDK37:14pass -> exact production revert13pass1fail -> exact restoration14pass; zero errors/skips.
The extended existing test uses a focused Activity/editor/InputConnection and registered framework alphabetic keyboard42. It checks both Return preferences, forwarded hardware Return/Ctrl/Shift chords, and unknown-origin IME newline/no-submit behavior. Test count unchanged11editor+3placeholder.
Fixed production SHA256 bd956781d8ebd651864adf085009e0f404fc78af0dc662e4d046804a69b1194e.
Frozen editor tests SHA25647b0a722fdac31ee61edecf2f45cce4f82a402da300fdb569bef6ad31f106a79.
Native test inputs byte-identical after main merge325a7e2. Postmerge46focusedTypeScript cases and mobiletypecheck pass. Ktlint1.7.1 passed scoped files.
Independent whole-diff Opus5.5 and GPT6Sol R8 found no confirmed code findings.
This is framework regression proof, not physical HID verification. Real forwarding IME/Chromebook and iOS permission/push device flows remain human gates. No merge readiness claim.