SAGE in motion

A real conversation with SAGE inside the product, on a real decision. SAGE has three bounded roles: Translation (converting user intent into structured system requests), Reasoning (drawing on cached state to answer questions without invoking the engine), Bounded Orchestration (executing operations within the Decision Policy's authorized limits). SAGE never decides. Every decision goes through the Decision Twin under the Decision Policy's authority.

Three depths of explanation organize every answer. Headline — the one-sentence answer. Decomposition — the breakdown by component, traced to the audit trail. Deep Lineage — the full audit trail back to source events and decisions. The user can ask for more depth at any point. The CFO who wants the headline takes the headline; the data science lead who wants the lineage drills down without leaving the conversation.

Root-cause analysis is mechanical, not generative. When a planner asks "why did margin drop on Wednesday," SAGE traverses the dependency graph backward — the margin drop traces to specific expedites, which trace to specific supplier slips, which trace to specific named risk events firing. Every link is real, drawn from State, not inferred from prose. SAGE cites the underlying audit trail and the planner can click into it.

The discipline is what makes SAGE safe to trust. SAGE refuses requests outside its bounded authority. SAGE flags uncertainty. SAGE says "I don't know" when the answer isn't in the audit trail. The website's SAGE follows the same discipline — it names what it's adapting and why; it refuses competitor characterization it can't substantiate; it routes to PULSE when the question is sized-to-you not theoretical.

A real conversation in the product runs roughly like this. A planner asks "why did Customer-Alpha's order shift to next week?" SAGE answers, drawing on the audit trail, the pegging chain, the named events from the past 24 hours, citing specific decisions and dollar impacts. The planner asks "could we have prevented it?" SAGE walks through the counterfactual — the EVW at the moment of the original decision, the resilience score, the alternative posture that would have held. The planner asks "recommend a posture change for next quarter." SAGE composes a draft policy variant within bounded orchestration and surfaces it on the Frontier as a candidate. It does not commit. The human commits, after reviewing the Delta PVA.