A certified decision is an AI output bound to the execution that produced it. The inputs that triggered it, the model and parameters that generated it, the tools that were invoked, and the conditions that surrounded it are all captured in a Certified Execution Record and sealed with a cryptographic hash. The result is a decision whose provenance can be independently confirmed and whose record cannot be silently altered.
Cite this definition
"A certified decision is an AI output bound to its full execution context within a Certified Execution Record, sealed with a cryptographic hash that enables independent verification and makes any post-hoc modification detectable."
, Certified Decisions, NexArt Glossary. https://nexart.io/glossary#definition
What makes a decision certified
An AI system produces an output. That output may be a recommendation, a classification, an action, or a structured response. On its own, the output is an assertion: the system says this is what it decided. There is no way to confirm the claim without trusting the operator.
A certified decision changes the trust model. It binds the output to the complete execution context that produced it and seals everything with a certificateHash. The decision, its inputs, and the conditions under which it was produced become a single, cryptographically linked artifact.
Anyone can then verify that the decision was produced by the stated model, with the stated inputs, under the stated conditions. Verification does not require an API key, an account, or trust in NexArt. It requires only the record.
Decisions are not just outputs
The distinction matters. An output is what the model returned. A decision is what the system chose to do, in a specific context, with specific constraints. The same model can produce different outputs under different conditions. Without the context, the output alone is incomplete.
- An output is a model response: text, structured data, or a function call.
- A decision is a model response together with the execution context that produced it: inputs, parameters, permissions, constraints, and tool availability.
- A certified decision is a decision sealed in a Certified Execution Record whose integrity can be independently confirmed.
This is the shift from "the system reported this" to "the system produced this, under these conditions, and the record is intact."
Integrity guarantees
A certified decision inherits the full integrity surface of the AI Execution Integrity specification:
- Tamper evidence. Any modification to the decision, its inputs, or its surrounding conditions invalidates the hash.
- Attestation. The NexArt attestation node can confirm the record and issue a signed receipt, providing third-party validation.
- Independent verification. Anyone can recompute the hash and confirm integrity without access to NexArt infrastructure.
- Deterministic formation. The same execution context always produces the same hash, regardless of environment or field ordering.
Important distinction: Certifying a decision proves that the record is authentic and unaltered. It does not certify that the decision was correct, ethical, or optimal. This is integrity, not endorsement.
The role of context in certified decisions
A decision without context is a data point. To understand why an AI system made a particular choice, you need to know the conditions it operated under: what tools it could access, what constraints bounded its behavior, what prior state it was working with.
Context signals capture these conditions as structured, hash-protected fields. When included in a certified decision, they provide a complete, verifiable picture of the decision environment, not just the decision itself.
This is especially important for agent execution workflows, where decisions build on each other across multiple steps and the operating environment evolves throughout the workflow.
When decision certification matters
- Regulated industries. Financial, healthcare, and legal domains where AI decisions carry material consequences and must be auditable.
- Autonomous agents. Multi-step agent workflows where each decision point needs a verifiable record.
- Third-party accountability. Demonstrating to customers, partners, or regulators what an AI system decided and under what conditions.
- Dispute resolution. Providing sealed evidence of what occurred when outcomes are contested.
- Governance and compliance. Meeting auditability requirements under ISO 42001, SOC 2, or similar frameworks.
How certified decisions work
Creating a certified decision follows the standard CER workflow:
- The AI execution occurs, producing a decision with full execution context.
- The execution layer captures inputs, parameters, outputs, tool calls, and context signals.
- All fields are canonicalized and sealed with a SHA-256 hash.
- The resulting Certified Execution Record is optionally submitted for attestation.
- The certified decision can be independently verified by anyone, at any time.
From the blog
Make your AI decisions verifiable
Every decision backed by a Certified Execution Record. Start in under five minutes.
Certified decisions are built on Certified Execution Records and the AI Execution Integrity specification. Learn about context signals for capturing the decision environment, or explore agent execution workflows where certified decisions are critical. For definitions of key terms, see the glossary.