The transcripts view derived each segment's visible state purely from the observe-layer modality state, so a segment that was transcribed but never thought rendered identically to a fully-processed one (false green). Add a think axis read from the think-layer fold (segment_fully_sensed + segment_fully_thought, keyed by segment key) and thread it through both the per-segment zoom pills and the server-side range rollup (worst-wins). A fully-sensed-but-not-thought segment now renders in a distinct "awaiting thinking" state. Read-only: the route helper annotates per-request response dicts only. Updates the transcripts API baselines to carry the new per-segment/range think field (20260304 has no health log, so all segments read "awaiting"). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>