Glossary

Term	Definition
Runtime Gateway	HTTP service that forwards model calls and records economics evidence.
Request Ledger	Append-only economic evidence store served through the investigation API.
Event	One recorded fact or normalized record. In Blackridge, usually an EconomicsEvent.
Economics Event	Canonical body-free record for one request/attempt.
Canonical Event	Normalized event shape used by reports regardless of runtime/import source.
Attempt	One try within a retry or fallback sequence.
Request ID	Unique request identifier.
Event ID	Unique event identifier.
Trace ID	W3C/OpenTelemetry trace identifier.
Span ID	W3C/OpenTelemetry span identifier.
Tenant	Top-level customer or organizational boundary.
Project	User concept; map to app, namespace, workflow, or metadata if needed.
Team	Owning team for spend.
Workflow	Related set of model calls for one business operation.
Agent	Logical autonomous or semi-autonomous actor in a workflow.
Provider	Upstream model service such as OpenAI, Anthropic, or fake.
Model	Provider model name requested or selected.
Route	Provider/model decision for a request.
Fallback	Alternate provider/model used after primary failure.
Retry	Repeated attempt for one logical operation.
Retry Amplification	Extra spend created by retries.
Fallback Tax	Extra cost created by fallback from expected primary route.
Failed Waste	Cost of failed model calls.
Abandoned Branch	Branch that cost money but did not contribute to final output.
Semantic Duplicate	Duplicate-like work identified by supplied semantic cluster metadata.
Cache Hit	Request served from cache.
Cache Miss	Request not served from cache.
Cache Miss Opportunity	Potential savings from cacheable duplicate traffic.
Prompt Tokens	Input tokens.
Completion Tokens	Output tokens.
Total Tokens	Input plus output plus provider-specific additional token classes when reported.
Token Economics	Blackridge report model for cost, tokens, waste, opportunity, and confidence.
Economic Forensics	Evidence-based reconstruction of AI spend behavior.
Attribution	Ownership metadata for spend.
Attribution Confidence	Per-field confidence: exact, strong, inferred, weak, unknown.
Unattributed Spend	Known spend without useful ownership metadata.
Unknown Spend	Spend that cannot be priced or classified from available data.
Pricing Snapshot	Fixed price table used for cost math.
Cost Micros	USD cost in millionths of a dollar.
Savings Estimate	Modeled or observed cost reduction figure.
Confidence Range	Low/high range used when savings are modeled or inferred.
Data Quality	Coverage/confidence metrics in reports.
Dropped Event	Event not persisted because sink/queue failed or was unavailable.
Evidence Export	Controlled export of investigation evidence for review or audit.
Raw Record	Original captured evidence before report-level aggregation.
Provenance	Evidence of where data came from and how it was transformed.
Lineage	Parent/child relationships between requests.
Workflow Trace	Trace of model and possibly tool work across a workflow.
OpenTelemetry	Standard telemetry framework for traces, metrics, logs.
GenAI Semantic Conventions	OpenTelemetry conventions for AI spans; Blackridge does not claim full coverage here.
AIProvider	Internal Go model for a provider resource; no CRD YAML installed here.
AIRoute	Internal Go model for routing policy; no CRD YAML installed here.
AIQuota	Internal Go model and runtime budget policy primitive; no CRD YAML installed here.
Policy	Rule/configuration used for route, cache, or budget behavior.
Guardrail	Runtime behavior that warns, downshifts, requires override, or explicitly denies.
Rate Limit	Request throughput limit. Blackridge has max in-flight request protection, not a full rate-limit product.
Budget Limit	Spend threshold for budget guard evaluation.
Fail Open	Continue inference when telemetry/event persistence fails.
Fail Closed	Fail request when configured event persistence condition cannot be met.
Design Partner Assessment	Guided assessment using a customer workflow and the investigation API.