Add jargon-sensitive readability (Dale-Chall, jargon rate, word familiarity); fix tooltip source term to match the edit

Together: low concurrency + patient retries to avoid 503s; friendlier busy-provider error

Drop gpt-oss (leaks reasoning over echo endpoint); bigger token budget for reasoning models; cap Together ablations for usable latency

Add Together.ai simplifier models (Qwen3.5-9B, Gemma-3n) with provider labels; fancier model dropdown

Tooltip: AoA before/after arrow + source term shown

Default to DeepSeek V4 Pro for reliable abbreviation simplification

Deploy MedPlain web app (engine, server, UI, data, pipeline_core)