Accessible Ghost card menus: Ember-layer key handlers or upstream Lexical changes?

Ghost 5.x replaced the Koenig editor with one built on Lexical, leaving card insertion and reordering dependent on both Ghost's Ember admin components and upstream editor behavior. The unresolved decision is whether to ship ARIA roles and key handlers directly in Ghost's Ember card menus, or to contribute focus and semantics fixes upstream to Lexical.

A local Ember patch is faster to test and release within Ghost's own cadence, but it risks divergence and rework as the Lexical layer evolves. An upstream contribution improves the shared editor core but is slower and outside Ghost's release control. Ghost's accessibility position targets WCAG 2.1 AA and acknowledges ongoing admin work, making the editor the primary unresolved surface rather than the reader-facing theme layer.

Because editor internals changed significantly between the Koenig/Slate era and the current Lexical-based editor, older bug reports may no longer apply. Any decision should be validated against the exact deployed Ghost version.

Which approach should Ghost prioritize for keyboard and screen-reader card operations: patching the Ember components locally, contributing upstream to Lexical, or a hybrid that keeps Ghost-specific chrome in Ember while pushing generic editor semantics into Lexical? How can teams verify that a chosen path does not create duplicate or conflicting focus management across the two layers?