2026-08-19 - Concept intake and roadmap resequence

Documentation only; no code touched.

Done. Fourteen decisions captured from a planning conversation (D028-D041), covering portable mode, configuration formats, the GUI, process architecture, and two standing directives. Roadmap resequenced to add phase 0c (structural seams), phase 2b (GUI shell and library), a stretch section, and a fuller phase 4. CLAUDE.md gained principles 11-13. ACKNOWLEDGEMENTS.md created.

The consequential one is D032 - the guest executes in a child process, not in the shim. The deciding argument is address space: the guest demands fixed addresses, and in a shim's process those compete with the UI toolkit, the graphics driver, loaded DLLs, and ASLR, so a load failure would be nondeterministic. Thread reclamation, clean restart, and fault-handler ownership all compound it. Cost is deferred to phase 6, where output produced in the worker has to reach a window in the shim.

Surprises.

• An initial in-process recommendation was a shortcut, caught only because the standing "no urgency, highest-payoff path" directive was stated explicitly. That directive is now D028 and CLAUDE.md principle 11 precisely because it changed an answer once already.
• The first thing that writes to disk is GUI settings at phase 2b, not traces at phase 4 - so orbistoun-paths had to move earlier than first planned. Worth re-checking that assumption whenever a new write appears.
• Batching decision-logging until "the end of planning" was the wrong call. Fifteen decisions accumulated in conversation before a prompt to verify caught it. Log as you go; the batch never feels urgent until it is lost.

Next. Unchanged - phase 0, with 0b and 0c available in parallel.