Most AI advice starts answering immediately. Decision Panels asks you the two or three questions whose answers would actually change the recommendation — then convenes a panel that argues it out and tells you how sure it is.
Anyone can ask a model for several opinions. What makes a panel useful is the order it does things in — calibrate first, reason independently before converging, and keep the disagreement instead of averaging it away.
Restate the decision in one sentence, name the options actually on the table — including "do nothing" — and establish how reversible it is. That sets how hard the panel should push.
The questions whose answers would most change the call, chosen for this decision: what winning means to you, the binding constraint, the private context no panel can guess. This is what makes a panel sharp instead of hedged.
Three lenses that disagree about most things give a one-line read each. If they land on the same call and the same crux, there is nothing for a panel to resolve — so it says so and stops, and tells you what would have made it convene. Hard-to-reverse decisions are never gated this way. Want the full panel regardless? Say so and it runs.
Every panelist takes a position before anyone converges — its recommendation, its reasoning, and the one assumption its view hinges on. No drifting toward each other.
Every take is stripped of its author and relabelled A–E, then each panelist reviews the others: which argument is strongest, which has the biggest blind spot, and what did all of us miss. Nobody knows who wrote what, so arguments get weighed on merit — and a panelist may only change position by naming the argument that moved it. Agreeing with the room is not a reason.
The most valuable output is the disagreement a single opinion would have hidden — stated at full strength on both sides, and only if it survived cross-examination. If the panel is nearly unanimous, it says so plainly.
Before the panel is convened, the skill takes a solo baseline — one strong single-pass answer — and hides it from the panel. At the end it reports whether the panel actually beat it, including the runs where it didn't: "the panel did not beat the single answer here." A panel that can't tell you when it was unnecessary isn't one you can trust when it says it was.
The method itself — runs the intake questions, convenes the panel, surfaces the disagreement steelmanned on both sides, and writes the brief. Install once in Claude Code, desktop, or web.
Lenses spanning founder decisions, product calls, advisory work and personal high-stakes choices — each with its own worldview and its own stated bias. The panel seats five to seven of them, starting with the ones that actually disagreed about your decision.
What makes every run land the same way: the ranked recommendation, the confidence level, the trade-offs, and the strongest objection — one page you can act on or forward to whoever else is deciding.
A real decision moving through the panel end to end, so you can see exactly what to expect — and reuse it as a starting point for your own.
A decision brief captures what you decided and what you assumed — on one particular Tuesday. Three months later the assumptions have moved and you no longer remember which ones the call rested on. That's the problem Chorusse Memory exists to solve: keep the brief, bring it back when it matters, and check it against what actually happened.
Nothing. Decision Panels is free — no account, no subscription, no upsell inside the tool. It runs on your own Claude, so the only thing it costs you is your own usage.
Yes. Decision Panels runs inside your own Claude — Claude Code, desktop, or web — on your own account. Nothing is routed through us and no part of your decision leaves your chat.
As a small set of files you drop into Claude as a Skill: the method, the lens library, the brief template and a worked example, with a short setup guide.
Entirely. There is no Chorusse server in the loop — we never see the question, the panel, or the brief. Your conversation stays between you and Claude.