Agent and Role Taxonomy

Concordia names roles by responsibility rather than by model provider.

Role	Responsibility	Authority limit
Rowan — Proposal Sentinel	Normalize and route the proposal	Cannot approve or execute
Mercer — Treasury Intelligence	Analyze allocation, liquidity, and evidence	Advisory only
Verity — Constitutional Risk	Challenge policy violations and preserve dissent	Cannot mutate the plan
Alden — Governance Strategist	Produce a bounded response plan	Plan requires deterministic checks
Locke — Casper Execution	Prepare the exact approved action	Cannot bypass human approval or hash binding
Wells — Governance Archivist	Summarize and publish evidence references	Cannot declare failed evidence valid
Concordia Core	Policy, integrity, approval, and state transitions	Deterministic code, not an agent persona

Card chain

Each role emits typed cards into a proposal-scoped chain. The chain records sequence number, prior-card identity, typed content, and publication state. Public views must preserve dissent and superseded evidence rather than silently rewriting the decision history.

Model boundary

Agents may suggest text or analysis. They do not own allocation arithmetic, approval validity, nonce consumption, proof verification, or transaction success. If a model answer contradicts deterministic policy, deterministic policy wins.

V1.5 boundary

This taxonomy describes the V1 council. Governance v3 is excluded from V1.5; historical parser names in the verifier do not add agents or runtime roles.