Publish Chimera GFX source

2026-09-03 03:27:14 +02:00
commit a6037502d7
828 changed files with 100454 additions and 0 deletions
@@ -0,0 +1,76 @@
+# Phase 1.0K write-diagnostic one-shot runner
+
+Status: **OFFLINE RUNNER PREPARED; NO DEVICE AUTHORIZATION**.
+
+Phase 1.0K adds host-only parsing and authorization support for the unchanged,
+ignored Phase-1.0J artifact. It creates no target artifact and performs no PS5
+connection, transfer, execution or result reception.
+
+## Exact binding
+
+| Object | Identity |
+|---|---|
+| Chimera GFX Phase J | `05d316e4d48660ab24d116153f094ef7e2f69522` |
+| RetroArch J artifact source | `2e385265ab893dbba5898178dde5e108d3dbfa19` |
+| RetroArch K host runner | `ee965a0be3cd3e0032330680e7614c766688410a` |
+| Runner source SHA-256 | `4ee58f08ff51cff3624cbc072c0e915e8c415eb8cd98185fa0d6a20c02b7c330` |
+| J artifact | 1,845,208 bytes; SHA-256 `6ff0f7ea391da5f15ea43512a871078133e896a6900ae9f8f3fa75711abb8009` |
+| J linker map | 637,728 bytes; SHA-256 `19f1cf851ad8f99d31d5de3a14591f81faa83589c18e93cd60be96ebef5d7719` |
+
+The runner change touches host code and tests only. The artifact source,
+artifact bytes, SDL overlay and target protocol emitter remain unchanged.
+
+## Wire contract
+
+J extends historical diagnostics without renumbering them. The parser therefore
+uses this exact 30-entry table:
+
+```text
+0..12   D00..D12
+13..27  I00..I14
+28      C1
+29      D13
+```
+
+Magic is `CHD10J01`, version is 1 and every frame is 64 bytes in big-endian
+wire order with CRC-32 over the first 60 bytes. A wrong table, including an
+apparently natural but incorrect `D13, C1` suffix, is rejected before socket
+creation.
+
+## Fail-closed activation contract
+
+The CLI has no protocol selector. J is selected only by an exact active
+`PHASE_1_0K_WRITE_DIAG_ONE_SHOT_RUNNER` manifest. Before transport the runner
+requires:
+
+- exact artifact name, profile, size and SHA-256;
+- `execution_eligible=true` and `transfer_eligible=true` in a separately
+  reviewed active manifest;
+- explicit one-shot connection, transfer, execution and reception authority;
+- a separate untracked approval for exact firmware 9.60, target, port, timeout
+  and unique run ID;
+- false installation, autoload, device-write, retry, reconnect, resume and
+  automatic-reboot fields;
+- exact agreement on magic, frame size and the complete wire table.
+
+The runner rehashes the artifact and durably creates an exclusive consumed-run
+receipt before any socket. It permits one connection, one `sendall`, one
+write-half-close and bounded reception. Trace creation is exclusive, capture
+is capped at 65,536 bytes, and there is no retry, reconnect, resume or
+overwrite.
+
+The tracked manifest and approval template are deliberately inactive: no
+target, port or run ID is present; every authorization and artifact eligibility
+field is false. These files cannot activate transport.
+
+## Offline evidence and limits
+
+Fake-socket tests cover fragmented J frames, C1/D13 index preservation,
+terminal D12, exact manifest selection, incorrect stage-table rejection,
+inactive-manifest rejection before transport, consumed-run protection and the
+inherited timeout/capture limits.
+
+This phase provides no firmware evidence. E118's historical operation, J
+runtime behavior, submit errno, visible presentation and complete cleanup all
+remain unproven. A future device action would require a new, exact,
+artifact-specific permission and a separately reviewed active record.