Ship v22.1.1: fast-mode <think> fix (Fix 3) + thinking on/off terseness split; archive v22.1

Upload plates/card_swe_sharp.png with huggingface_hub

Upload README.md with huggingface_hub

Upload plates/card_swe_sharp.png with huggingface_hub

Upload README.md with huggingface_hub

Upload plates/card_swe_bounds.png with huggingface_hub

Rebase Sharp onto froggeric v22.1 (terseness-only, medium default); archive v22-sharp

Claw-Eval plate: 'Better answers, fewer words'

Add Claw-Eval output-changes plate + README fixes (alt-text, wording)

Add MMLU-Pro header plate (stock vs Sharp Qwen3.8-27B) + v1 archive; proofread README

Sharp v22: suppress forced-xhigh reasoning default (opt-in preserved via chat_template_kwargs); archive v1; card: shipped-models point to v1, add 'Setting reasoning effort' section

Add oMLX and llama.cpp runtime install paths, both verified; correct the precedence warning

Fix gguf flag, add oMLX precedence warning, multi-turn caveat, check_applied.py

Card: use % throughout

Card: colour the stock-vs-Sharp deltas

Sharp: froggeric v21.3 + terseness prompt, model-agnostic

initial commit