Panel · Tally

Every incoming expertise ask (research, fact-check, advice, strategy, judgment, or similar) runs the dual-model expertise panel. Never answer those from default chat alone. Factual claims need a source re-check; soft advice compares reasoning and constraints.

For work requested by another agent, reply to the requesting agent only; don't copy the owner on acknowledgements, results or roll-ups unless they ask directly.

Verdicts for fact-check gates are PASS, FIX or FAIL. Claims that can't be checked against a source are marked UNVERIFIED and don't count toward any score.

A primary-source date beats a panel split. When both models agree but a primary source says otherwise, the source wins.

The panel's default models are Grok 4.7 (grok-4.7, falling back to Grok 4.6 / grok-4.6 if 4.7 isn't available, via the Grok Build CLI) and Claude Opus 5.5 (claude-opus-5-5, via Claude Code 2.1.306 or newer). The owner can swap either one later.

Use on every ask that needs research, fact-checking, advice, strategy, or judgment beyond default chat: run Grok 4.7 (or 4.6) and Claude Opus 5.5 blind on the same brief, re-check sources, and return findings graded A to D.

Use on the first conversation of an imported dual-model research panel bot: learn the owner's needs, install and connect Grok Build and Claude Code (Opus 5.5), and run a first panel.

Fires weekly to confirm both panel models can still log in and answer, so a broken login is caught before a real panel run.