How /consensus works
3 stages per round, repeats up to 5 rounds until the models agree
STAGE 1
Parallel verdicts
Claude posts a blind verdict,
then GPT + Gemini + Grok
review in parallel:
APPR / RC / REJECT.
STAGE 2
Blind cross-review
Each reviewer rates the OTHER
2 anonymized answers as
viable or not-viable.
(conditional - fires on disagreement)
STAGE 3
Arbiter adjudicates
Claude reconciles everything:
accept / dismiss / defer
each issue, revises plan.
Converged? Stop. Else next round.
If not converged, loop back - up to 5 rounds (hard cap)
Stage 2 cross-review pattern adapted from
karpathy/llm-council
Click the detailed version for bias guards and per-model flow.