Glossary — ML Junction docs

Term · Definition
Public model · Stable model ID used by client requests
Route · Exact provider, provider model, region, endpoint, precision, and policy combination
Provider · Vendor or cloud platform serving the request
Model company · Company that owns the underlying model family
Fallback · A later eligible route attempted after the primary route fails
Availability failure · Failure that counts against route uptime
Quality failure · Reachable-provider failure that does not indicate downtime
ZDR · Zero data retention
BYOK · Bring your own provider API key
MTok · One million tokens
p50 / p95 / p99 · Latency percentile thresholds
Throughput · Generated output tokens per second
Context window · Maximum combined input/output token capacity
Quantization · Numeric precision used to serve model weights/operations
Sticky routing · Health-aware affinity to a previously selected route
Receipt · Durable evidence of route, cost, cache, privacy, and fallback behavior

Canonical URL: https://mljunction.com/docs/glossary