max / makenotwork

Delete the QUASI_SCREENS flag: the description is the only rendering All seven converted screens serve from the description layer on both boxes, so the switch has nothing left to switch. A flag that outlives its conversion is where the second rendering path lives, which is the third root cause named in the frontend-unification note: a path nobody runs is not maintained, and the hand-written path beside it keeps growing. Gone: QuasiScreens itself, the config field, the startup warning, the .env.example block, quasi::described, six conditional route registrations, the project_content write guard, seven Askama handlers, build_ssh_keys, seven template structs and their .html, and the six parse tests in parity_harness. quasi::mounts is a vec! literal now: a screen missing from it is an address nothing answers. Two predicates read `screen.is_some_and(|n| screens.enabled(n))` and simplify in opposite directions. In settings_tabs::shown_at it kills a path -- a described section can never be selected, which is what made the "ssh-keys" match arm unreachable and build_ssh_keys dead. In user_tabs::html it keeps one: not every tab is described, so only the second half of the test retires. ProjectContentTabTemplate stops deriving Template. Both callers of build_content take items, deleted_items and posts off it and hand them to the described renderer; nothing rendered project_content.html any more. Its blog table went with it, which is what 6077d0d9 was waiting for -- quasi's project_content renders that table itself. Two seals move, in opposite directions. handwritten_confirm 47 -> 46: library_contacts.html took its hx-confirm with it, as Shape 8 predicted. The other two dual-serve templates stay, because they are not tab renderings but the fragments routes::api::ssh_keys and the git-token routes answer with. Dead vocabulary 44 -> 45, and this one loosens: .form-label lost its last spelling in this repo's markup, while makeover-webview still writes it on the screen that replaced that template. The strip tests narrow rather than go. A described tab now never carries hx-target, where under the old default every tab was Askama and every tab was retargeted, so each test keeps the Askama tabs and asserts the described ones get a frame but no target.