Where this actually stands (2026-08-24)

Phase	State
0 - synthetic fixtures	not done, and no longer blocking. Real material arrived first (D050), so this narrowed to the malformed cases a compiler never emits
0b - ABI spike	done, both platforms. Grew into orbistoun-thunk
0c - structural seams	done. cargo tree shows orbistoun-gpu with no path to ash, and the shims hold no logic
0d - test-corpus tooling	not done. There is no corpus sync; the corpus is material dropped in by hand
0e - observability substrate	done. Run reports, retention, and the diff-against-previous machinery
1 - container wrapper and dynamic segment	done
1b - corpus-wide survey report	done. ./bin/orbistoun sweep and orbistoun-cli worklist
2 - symbol resolution	done. Both halves: a database loads from disk, and the generator confirms names by collision
2b - GUI shell and library	done; no output surface
3 - address space	done, both platforms verified
4 - worker, placement, relocation, protection, stubs, entry	done
5 - threading and synchronisation	begun, and nowhere near its own observable result
6 - first pixel	not started; its contents are being built ahead of it

The honest reading of that table. Phase 5 is listed as current and is not what the work is actually about: mutexes and semaphores are built and exercised, but scePthreadCreate has never been called by any title, so every guest is still in single-threaded startup. The three walls in PROJECT_STATUS.md are phase 4 completion problems - getting one guest through startup - and until one of them falls, phase 5 has nothing to demonstrate against.