Agent Contract Data Modeling
Design-time contracts for agent-safe enterprise data products.
ACDM defines how agent-facing entities declare trust, freshness, provenance, cost, and risk before autonomous or semi-autonomous AI systems consume data and act on it.
Why ACDM exists
Agentic AI changes the consumer of enterprise data.
Traditional schemas and catalogs describe what data looks like, but they do not always tell an agent whether data is fresh enough, traceable enough, trusted enough, affordable enough, or safe enough for a specific class of action. ACDM makes those expectations explicit before runtime.
Methodology
The five agent-facing properties.
Trust
Declares confidence in the entity and how conservative downstream use should be.
Freshness
Sets maximum data age expectations so stale data does not silently drive agent decisions.
Provenance
Requires traceability to origins, transformations, ownership, and audit context.
Cost
Makes operational or monetary consumption costs visible before automated workflows run.
Risk
Classifies sensitivity and action risk so unsafe use is blocked or constrained.
Reference implementation
Validate ACDM contracts before agents consume the data.
The acdm-validator project turns the methodology into governance-as-code checks for entity contracts, composition rules, permission ceilings, maturity stages, and audit records.