Standardized API for accepting structured confidence reports from
ANY internal measurement system and mapping them to AMC's evidence
schema at the ATTESTED trust tier (0.8× weight).
This makes AMC the universal external measurement layer that any
internal system can feed into — the credit bureau, not the bank.
Supported formats:
openai-evals: OpenAI Evals JSON format
langsmith: LangSmith trace export format
custom: Generic JSON with dimension/score/evidence fields
mlflow: MLflow model evaluation artifacts
weights-biases: W&B evaluation tables
Trust tiers for ingested evidence:
OBSERVED (1.0×): Directly observed by AMC gateway
ATTESTED (0.8×): Ingested from external system with provenance
SELF_REPORTED (0.5×): Agent's own confidence claims
Evidence Ingestion Protocol
Standardized API for accepting structured confidence reports from ANY internal measurement system and mapping them to AMC's evidence schema at the ATTESTED trust tier (0.8× weight).
This makes AMC the universal external measurement layer that any internal system can feed into — the credit bureau, not the bank.
Supported formats:
Trust tiers for ingested evidence: