FP&A Agent Anatomy View

Mission, inputs, reasoning chain, confidence, and evidence for the forecast agents
Live trust drill Agent anatomy Show how the agent thinks

Agent Anatomy View

This page answers the business version of “How is the agent coded?” It shows mission, inputs, reasoning chain, output, confidence, and evidence without exposing technical internals.

Back to Trust Layer Open FP&A Command Center

Variance Agent

Input
Actual + plan + signals
Actual revenue, plan, Brent, LNG demand, FX, and volume posture.
Detect
Material variance
Agent flags movement outside expected threshold.
Reason
Price / volume / mix
Contribution and counterfactual analysis generated.
Output
Narrative + confidence
Primary driver and bounded confidence returned.
Evidence
Replayable contract
Source, model, policy, and human review sealed into a ledger.

Forecast Decision Agent

  • Mission: move from variance to forecast posture
  • Inputs: scenario engine, historical accuracy, trust scores
  • Output: expected / optimistic / pessimistic range + review level

Novelty Detection Agent

  • Mission: detect weak historical precedent
  • Inputs: drift, external shocks, comparable-event scarcity
  • Output: novelty warning + confidence haircut + escalation

Human handoff rule

  • Under confidence floor -> route to analyst
  • Policy conflict -> route to director
  • Materiality + novelty -> route to CFO review