The Deliberation Index tracks consensus rates, dissent levels, and confidence grade distributions across domains — as the corpus grows. Not what the models think. What the structured deliberation produces.
Consensus rate is the proportion of sessions in a domain that reached a Probable or Established confidence grade without an Unresolved Strong Contrarian objection. High consensus means the ensemble consistently converged. Low consensus means the questions in that domain consistently produce unresolved adversarial pressure — which is information, not a failure.
Cartographer, Methodologist, and Synthesizer converge. Contrarian objections are resolved or graded Speculative. Final grade: Probable or Established.
Contrarian raises Unresolved Strong objections that carry to Phase 3. The finding reflects the dispute — Contested grade or multiple unresolved objections in the output.
Each cell shows the average metric for that domain × confidence grade combination. High consensus doesn't always mean low dissent — a domain can produce many Probable findings while still generating significant Contrarian pressure on individual claims.
The preponderance of Probable grades reflects an accurate picture of where most research questions sit: supported by evidence, but not at the level of independent replication across methodological contexts. Contested at 31% is structurally honest — the ensemble isn't softening findings. Gap at 7% is small but important: these are the questions the evidence genuinely cannot answer.
AI governance questions produce a Contested or Gap grade in 71% of sessions — the highest contested rate of any domain. Technology is second at 58%. Life sciences and economics are the most resolvable domains in the corpus, with Probable or Established grades in over 80% of sessions.
A Contested finding is not a failure to reach a conclusion. It is a structurally complete finding that names the dispute, preserves competing positions, and surfaces the resolution condition. In 31% of all sessions, the honest answer is "the evidence is genuinely split." The Index records that accurately.
The Index is a descriptive record of how the ensemble behaves across domains — how often it converges, how often the Contrarian's objections go unresolved, and how findings distribute across the four confidence grades. It measures the epistemic behaviour of the system, not a claim of predictive accuracy.
The most useful signal is where deliberation breaks down. A high Contested-or-Gap rate in a domain is not a weakness in the system — it is an accurate reading that the evidence there is genuinely unsettled. The ensemble is built to say so rather than manufacture confidence.
Browse every session by domain, confidence grade, dissent level, and session depth. Sort by domain, grade, or recency.
Open explorer →Interactive heatmaps showing how consensus rates, dissent levels, and confidence grade distributions vary across domains and evolve over time as the corpus grows.
View heatmaps →Full methodology — how consensus and dissent are measured, how confidence grades are assigned, and what the data does and doesn't show.
Read methodology →Every session you run contributes to the corpus.
Join the waitlist — one Standard session free.