Publish Chimera GFX source

docs/retroarch/phase-1.0o-write-free-device-result.md

@@ -0,0 +1,53 @@
+# Phase 1.0O write-free one-shot device result
+
+The exact Phase-1.0M artifact was transferred and executed once on firmware
+9.60 through the separately authorized one-shot N runner. The attempt is
+consumed. There was no installation, staging, autoload, device filesystem
+write, retry, reconnect or reboot.
+
+## Result
+
+The source correction worked: valid frames reached I04 and all later startup
+intervals through I14. No D13 was emitted, so this run observed no write
+firewall activation. RetroArch entered the SDL video driver, opened VideoOut
+with positive handle `0x4e100100`, and recorded buffer registration result
+`0`.
+
+The early diagnostic copied its pattern and called
+`sceVideoOutSubmitFlip(handle, 0, 1, 0)`. D07 recorded submit result `-1` and
+the immediately saved `errno` value `0`; the event wait was therefore not
+called. The SDL overlay then followed its generic `framebuffer_fail` path,
+emitting D12 `(RUNTIME_FAILURE=5, E104_FRAMEBUFFER=104)`, and SDL initialization
+returned `-1` in D04. E104 does not prove an allocation failure here: source
+order binds it to the generic label used after the preceding early-flip
+failure.
+
+This reproduces the Phase-1.0H flip boundary after removing the Phase-1.0K
+`MKDIR` blocker. It strengthens `sceVideoOutSubmitFlip()` as the current
+source-bound failure site. It does not explain why that public driver call
+returns `-1`, and `errno=0` supplies no additional error identity.
+
+## Transport evidence
+
+- artifact: 1,845,208 bytes; SHA-256
+  `c99a0856309a357ad2667d89b4924e4063ad214cae09c8a419457b0732f583cd`;
+- one connection, one `sendall`, one write-half-close;
+- 3,953 bytes received in 14 reads, followed by remote EOF;
+- 25 CRC-valid frames, no parser errors and no capture truncation;
+- trace: 18,031 bytes; SHA-256
+  `3d0b8811ae11f5cac2c2d331e252e1789a10588cab0e6045828a6b6af0fe1eb6`;
+- zero retry and zero reconnect.
+
+The runner returned its strict incomplete status because D12 carried the
+terminal flag at sequence 24 while D04 followed at sequence 25 in the same
+received block. Consequently the last parsed frame was nonterminal. This does
+not invalidate the individual CRC-checked frames or lossless stdout, but it
+does mean a successful terminal/cleanup contract was not observed.
+
+## Remaining boundary
+
+Visible presentation, successful flip submission, wait-event behavior,
+complete cleanup and safe process exit remain unproven. All current
+authorizations and artifact eligibility fields are false. The local active
+manifest, approval, receipt and raw trace remain ignored evidence and are not
+committed.