Publish Chimera GFX source

docs/retroarch/phase-1.0p-terminal-ordering-analysis.md

@@ -0,0 +1,24 @@
+# Phase 1.0P terminal ordering analysis
+
+The Phase-1.0O order `D07, D12, D04` is deterministic in the bound source; it
+is not evidence of network reordering.
+
+1. The early diagnostic detects the nonzero submit result and enters SDL's
+   `framebuffer_fail` path.
+2. `chimera_ps5_smoke_video_error(E104)` requests failure shutdown.
+3. The shutdown request synchronously emits D12, which the current stream
+   helper marks terminal.
+4. `PS5_VideoInit` then returns failure to `sdl2_gfx.c`.
+5. The caller emits D04 with the SDL initialization result `-1`.
+
+D12 and D04 arrived in the same host receive block. The runner intentionally
+requires the last parsed frame to be terminal, so D04 correctly prevented a
+successful terminal classification. The individual CRC-valid frames remain
+runtime evidence; successful cleanup and safe exit do not.
+
+Accepting any earlier terminal frame would incorrectly promote a shutdown
+request into lifecycle completion. A future protocol must instead use a new
+magic and a distinct final stage emitted only after D04 and bounded cleanup or
+return evidence. D12 may remain failure/shutdown-request evidence, but must not
+serve as the lifecycle terminal on this error path. This is a design
+requirement only: Phase 1.0P changes neither target code nor the host parser.