What Is Included in an InferTrust™ Decision Record?

Every InferTrust™ decision record captures model version, confidence score, policy hash, input feature hash, timestamp, decision action, and sequence ID in one

What Is Included in an InferTrust™ Decision Record?

Every InferTrust™ (Patent Pending) decision record captures a comprehensive snapshot of the AI inference event, sealed with a cryptographic signature to prevent post-decision modification. The record includes the following components:

Core Decision Data

• Input reference: a cryptographic hash of the input data the model evaluated, linking the decision to its evidentiary basis
• Model output: the raw output produced by the AI model before any post-processing
• Confidence score: the model's confidence level for the specific decision
• Routing decision: whether the decision was approved autonomously or escalated for human review

Model Identification

• Model version fingerprint: a cryptographic hash uniquely identifying the model's weights and architecture
• Configuration parameters: inference settings such as temperature, sampling parameters,and thresholds
• Training metadata reference: a link to the training run and dataset that produced the model

Integrity Metadata

• Trusted timestamp: when the decision was made, sourced from a trusted time authority or local hardware clock
• Hash chain reference: a link to the previous decision record establishing temporal ordering
• Digital signature: the cryptographic signature sealing the entire record against modification
• Signing key identifier: identifying which key produced the signature for verification purposes

This comprehensive record ensures that every element needed to reconstruct, verify,and defend the AI decision is captured and preserved in a tamper-evident format. Learn more about the InferTrust™ platform and its approach to verifiable AI evidence.