CONFIDENCE & TRANSPARENCY
Every other AI gives you an answer. Counsel gives you an answer — and shows you where the models agreed, where they diverged, and why the synthesis chose what it chose.
See it in your own queries View benchmarks →
A council response is one answer accompanied by three pieces of metadata other AI APIs don't surface:
4/4 agree — high confidence. The synthesis is strong. Act on it.
3/4 agree, 1 dissents — the dissent might contain the most important nuance. The synthesis lifts it into a named caveat.
2/2 split — a genuinely contested question. The synthesis names the trade-off; you make the call.
Agreement is confidence. Disagreement is a caveat worth naming. A single model gives you the answer; a council gives you the answer plus its own measure of certainty.
The live page includes three example queries you can step through — a system-architecture question (3/4 + caveat), a legal question (3/4 strong support), and an under-specified question where the council does not converge and the synthesis becomes a decision tree. Try them in your own dashboard after signing up.
It reflects the share of council members whose response converges on the synthesised answer. 4/4 means full agreement; lower percentages name where the dissent lies.
Each council response includes the individual model answers, the anonymous peer review scores, and a per-model attribution of which views shaped the synthesis.
When agreement is in the 50-80% band. The dissent usually contains the most operative caveat — a specific edge case other models missed.
Yes. They report intra-council convergence, not ground truth. A 100%-agreement answer can still be wrong if every model shares the same blind spot. That is why we ship the reasoning trace alongside.