v1.0.0 - honest demotion to research PoC

v1.0.0 - honest demotion to research proof-of-concept

This release is an honest repackaging, not a capability announcement.

What changed

• PROJECT_STATUS.md added at repo root: states plainly that ARC-AGI accuracy is ~2-5% on 100-task subsets, the symbolic reasoning engine does not yet perform, and this is NOT an ARC Prize submission.
• README rewritten to remove "solves ARC" / "AGI architecture" / "submission-ready" language. New framing: GPU compute-shader playground.
• Installable package arc2-chimera (Apache-2.0) exposing the reusable shader primitives: matmul, GELU, diffusion, cellular automaton, holographic memory.
• CLI: arc2-chimera info (prints disclaimer), arc2-chimera evaluate (runs one ARC task; falls back to an identity baseline if the GPU engine is not available, rather than pretending to solve).
• Offline pytest suite (12 tests, all passing).
• CHIMERA/arc_chimera_submission.py moved to archive/submission_unverified.py with a NOT VALID FOR ARC PRIZE 2025 banner.
• Windows cp1252 UnicodeEncodeError fixed in existing test.

What did NOT change

• The symbolic reasoning engine's accuracy. It is still ~2-5%.
• Existing code paths in CHIMERA_ARC_OpenGL v5-v10, CHIMERA_ENIGMA, CHIMERA_RIVER. They are kept for historical reference.

Reusable value

The GLSL compute kernels and the GPU-texture state loop. That part works.