Bench on the live demo: Clef-flash 9/12 right at ~0.5 s per decision, confidence not usable as a gate (right 0.04–0.39, wrong 0.12–0.28); Llama 3.3 fast 12/12 at median 2.6 s. So: brain="fast" on <ai-chart> draws Clef's chart immediately and asks the LLM in parallel; the chart is redrawn only if the LLM's spec differs (the line under the chart says first draw / LLM agrees / corrected). The series and single-value filter questions are reworded around the three misses. Server default for API callers stays llm.
Bench on the live demo: Clef-flash 9/12 right at ~0.5 s per decision, confidence not usable as a gate (right 0.04–0.39, wrong 0.12–0.28); Llama 3.3 fast 12/12 at median 2.6 s. So: `brain="fast"` on `<ai-chart>` draws Clef's chart immediately and asks the LLM in parallel; the chart is redrawn only if the LLM's spec differs (the line under the chart says first draw / LLM agrees / corrected). The series and single-value filter questions are reworded around the three misses. Server default for API callers stays `llm`.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
https://claude.ai/code/session_01VDkN7ukK1kXkqov8FaZxva
---
_Generated by [Claude Code](https://claude.ai/code/session_01VDkN7ukK1kXkqov8FaZxva)_
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Bench on the live demo: Clef-flash 9/12 right at ~0.5 s per decision, confidence not usable as a gate (right 0.04–0.39, wrong 0.12–0.28); Llama 3.3 fast 12/12 at median 2.6 s. So:
brain="fast"on<ai-chart>draws Clef's chart immediately and asks the LLM in parallel; the chart is redrawn only if the LLM's spec differs (the line under the chart says first draw / LLM agrees / corrected). The series and single-value filter questions are reworded around the three misses. Server default for API callers staysllm.🤖 Generated with Claude Code
https://claude.ai/code/session_01VDkN7ukK1kXkqov8FaZxva
Generated by Claude Code