Usage Record

Usage records exist only as raw log lines buried in application infrastructure. When the billing team asks 'how many API calls did Acme Corp make last month?', an engineer SSH's into a server and runs a grep command. Nobody has formally defined what constitutes a billable usage event versus an internal health check.

None — AI cannot identify or classify billable usage events because no formal definition of a usage record exists. Metered consumption is invisible to any automated system.

The finance team has a wiki page titled 'Usage-Based Billing Overview' that describes metered events in prose: 'API calls are counted per customer per month.' But the definition doesn't specify whether retries count, whether webhook callbacks are included, or how partial-second timestamps round. Engineers interpret the rules differently depending on who wrote the logging code.

AI can surface the wiki page, but cannot resolve ambiguities in the usage record definition or determine which raw events qualify as billable without human judgment on edge cases.

Usage records are documented in a standard specification that lists required fields: customer ID, metric name, quantity, timestamp, and event source. The team maintains a 'Metering Playbook' in Confluence that defines each billable metric, its unit of measure, and aggregation rules. New metrics go through a review process before being added.

AI can reference the metering specification to validate incoming events, but cannot programmatically enforce field completeness or detect schema drift without structured machine-readable definitions.

Every usage record conforms to a versioned schema in the company's schema registry: customer_id (UUID), metric_name (enum), quantity (decimal), timestamp (ISO 8601), idempotency_key (string). The billing team can query 'show me the schema for compute-hour usage records as of January 1st' and get a precise, timestamped answer. Schema changes go through a formal RFC process.

AI can validate every incoming usage event against the registered schema, flag malformed records in real time, and generate billing calculations directly from schema-defined aggregation rules.

Usage records are nodes in a formal billing ontology. Each metered event schema links to the pricing tier that consumes it, the contract clause that authorizes it, and the invoice line item it generates. An AI agent can answer 'which usage record fields feed into Acme Corp's overage charges for API calls?' by traversing the ontology without human guidance.

AI can autonomously trace any usage event from raw ingestion through aggregation, rating, and invoicing — identifying pricing errors, contract mismatches, and revenue leakage across the entire billing chain.

Usage record definitions are self-documenting: the metering SDK auto-generates schema documentation from code annotations, the pricing engine publishes its consumption rules as machine-readable specs, and contract management emits entitlement schemas. When a developer adds a new metered feature, the usage record specification, pricing rules, and billing documentation update simultaneously without human authoring.

Can autonomously maintain the complete usage record specification lifecycle — from definition through validation, pricing integration, and audit documentation — without human intervention.

Ceiling of the CMC framework for this dimension.