REPL architecture direction

A design study of where the REPL should go. Three architectures were weighed:

• a block notebook with persistent-turn semantics and structured cards;
• a stream console extending the current line-editing substrate;
• a hybrid workbench combining a shell transcript with dockable panes.

The recorded recommendation is sequenced:

• stream console near-term, since the current codebase already does the line-editing part well enough to serve as the substrate;
• hybrid workbench long-term, built on full-screen / pane machinery (Ratatui + Crossterm).

A guiding observation: ral’s top-level turns are semantically special — they carry persistent state — and the REPL should make that visible rather than hide it.

This is a direction, not yet a committed implementation. The stream-console substrate is in place — rustyline line-editing over a scrollback transcript — and the REPL’s editor builtins already live in the right layer (repl-builtins-stay-in-repl). The Ratatui + Crossterm stack the long-term workbench would build on is now exercised in-repo by exarch’s full-screen TUI, though ral’s REPL has not adopted it.

See also core.