Google-side LLM stack updates

A background-noise week in the Google-adjacent LLM stack. Ornith-1.0, a self-scaffolding LLM harness for coding agents, took the anchor slot, while arxiv work on LoRA variants for continual motion-language agents ran in parallel. No headline model launch, but a tightening pattern: reuse the base, wrap it in cheap adapters, keep old skills around. The cluster's alignment to the Google-infra topic is soft, so this reads as adjacent movement rather than a clean flagship story. Watch for a stronger anchor next week.