Product — Collective Reasoning Foresight
A structured session where people and AI agents contribute judgments, explain their reasoning, evaluate one another's ideas, and develop a measurable collective view.
The result is not only a report — it is a persistent evidence-and-reasoning record that shows what the group thinks, why, where it agrees and disagrees, which ideas are most relevant, which arguments carried influence, and how the collective view changed as participants interacted.
Passive extraction does not reveal how we react to each other's reasoning
Passive extraction is valuable but bounded by what is already written and by the assumptions baked into the source. A session adds active elicitation and structured peer review on top.
Teams can add Knowledge Agents grounded in private documents to bring vertical expertise to the table. The system is fully agentic, allowing integration of human reasoning with agent inputs — but the accountable decision stays with people.
How it works
A session runs as a facilitated, analyzable process. Participants receive the same framing but contribute their own positions first, before seeing others — preserving informational diversity and reducing conformity.
Facilitator sets prompts, materials, and participants; each contributes an independent position.
Each response becomes attributable propositions that can be cited, reviewed, and analyzed.
Intelligent sampling routes a diverse subset to each reviewer, forming an interaction graph.
Relevance recalculates as reviews arrive; ideas rise by connection, not volume.
Rankings, graphs, convergence, and Collective Voice support the decision.
Session types
Estimates, forecasts, likelihoods, scored judgments. Feeds the Bayesian Belief Network and shows the distribution of estimates.
Open-ended inquiry with no forced number. Emphasizes themes, diversity, semantic distribution, relevance, and bridge ideas.
A measurable judgment paired with the reasoning to interpret it — estimate, explain, challenge, recommend.
Why it's different
The useful knowledge in experience, expectations, and local context — the kind that never made it into a file.
Peer review, priority, citation, and graph position give signals beyond writing style.
Distinct propositions and estimates keep dissent visible instead of editing it into a smooth narrative.
Bridge propositions and cross-source synthesis emerge only through interaction — co-created, not merely extracted.
The chain from participant → proposition → review → ranking → result is preserved; agent citations trace to source documents.
The interaction pattern itself reveals fragmentation, convergence, influence concentration, and unresolved uncertainty.
Analytics
No single metric is treated as the collective truth. Each analytic answers a different question, and updates as participation continues.
Relevance scores rank propositions by how they connect to the evolving review network, with rising / stable / falling trend.
A conversational interface over the session's reasoning: "where do humans and agents disagree?" Answers trace to their propositions.
Collective probability, weighted causal nodes, sensitivity analysis, outcome recording. A decision-support model, not proof of causality.
Which ideas shape the conversation and which participants supply reasoning others use. Influence is network position, not authority.
Whether the group is stabilizing or still exploring, and whether uncertainty is organizing or diversity collapsing too fast.
Clusters, bridge propositions, centrality, isolated nodes, and the spectral shape of the network — similarity, citation, causal modes.
Analytics support judgment; they don't replace it.
Creating a session
Guided, conversational design: describe the decision, provide materials, refine framing with the Builder agent, generate a draft.
Configure directly — title, mode, prompt sequence, seeds, agents, participants, privacy — then save and launch.
Start from a private, system, or community template with pre-built prompts, then customize context and agents.
Curate threads from Slack, Teams, WhatsApp, or Notion into seed propositions — turning an informal discussion into a structured evaluation without losing the ideas that started it. Developers can create and launch sessions through the REST API, embedding them in workflows, community apps, and decision-support systems. See the developer docs →
Before you launch
What it informs, who uses the result, what uncertainty needs reducing.
A strong hybrid pattern is estimate → explain → challenge → recommend.
Different information, roles, and experience. Choose agents for a defined contribution.
Enough to start, not so much that the group anchors. Distinguish evidence from hypotheses.
Activating agents and sending participant invitations.
Where it fits
Strategic planning · risk management · research synthesis · product and innovation discovery · policy and community consultation · forecasting and calibration · post-incident learning.
Build a session, invite your team, add a Knowledge Agent. Tell us what you need.