Changelog
Every release since launch: kernel updates, detection-vector patches, and product changes, with date and scope. Proof we're still shipping.
Keep Error Lens cards horizontally stable while scrolling code.
Six-model Solve lineup, live usage meter, and streamlined Panda chat.
Chat UX overhaul: top-anchored reading with a reader-held scroll, collapsed-bubble recall, and a calmer attention model.
Voice smart answer + recap lanes upgraded to GPT-5.6 Terra (shadow-eval winner: better answers, ~2.4x faster, cheaper).
Q1 deep-arbiter precedence for updated findings + Q2 multi-line error-lens (spans, multi-row diff card, narrowest-wins overlap).
Error-lens: findings shown on the editor (severity squiggles + diff card at the flagged line, not just the rail). Live utterance chat ordering + agent-gated keep; reviewer truthfulness + dedupe hardening.
Live utterance zone: kept bubbles now follow the assist gate's real judgment (agent-officialized history); live-voice mint backoff + STT contextual priming.
Live utterance zone in chat (tentative + kept bubbles, real STT), per-interview-type transcription tuning, smoother live-voice reliability (mint backoff).
Live transcription indicator: a subtle activity trace by the Listening pill shows the words Aceloop is hearing in real time (interviewer and you in distinct colors).
Reliability and polish pass.