max / quasi

0.52.0: a box with the focus keeps the letters a binding would have taken bc5528e2, found 2026-08-16 porting audiofiles' help overlay -- the first consumer Chrome ever had. Runtime::pressed_binding read raw input at the top of show, before the screen drew, and matched every binding with no focus guard. Sixteen of audiofiles' twenty-six shortcuts are bare letters, and declaring any of them meant the tag field, the rename pattern and the search box stopped accepting that letter. The port declared four -- f1, f2, ctrl+t, ctrl+shift+m -- and left the rest unsaid. The guard is the renderer's, as the finding argued: which keys a box eats is a fact about this keyboard and this host, so a Binding that answered it would be answering for hosts it has never seen. ctx.text_edit_focused() plus types(), which is narrower than the shipped app's rule in both directions -- that one suppresses everything while anything has focus, so it kills ctrl+t mid-word and fires on a button reached by Tab. Matching goes exact in the same pass, and it is a defect the guard exposed rather than one it caused. matches_logically ignores a shift the pattern never asked for, which is right for one control's Act::key and wrong for a table: an app binding both f and shift+f had the bare one answer both, first match wins. Nobody could see it while bare letters were undeclarable. Three tests, one driving two frames through one context so the focus under test is real rather than stubbed. Consumers forward-fixed in the same pass: audiofiles takes it and its chrome goes from four rows to thirteen; MNW server, goingson and the scaffolder's template move to 0.52.