Glossary
41 definitions in plain language. Each entry is a direct answer, not an essay.
A
B
C
- Canary ReleaseReleasing a new version to a small slice of traffic or data first, comparing health against the stable version before full rollout.
- Change ManagementThe process by which modifications to systems are proposed, assessed, approved, scheduled and recorded.
- ChunkingSplitting documents into retrieval units with metadata, balancing semantic completeness against retrieval precision.
- Circuit BreakerA pattern that stops calling a failing dependency after a threshold, failing fast until it recovers, so the failure doesn't cascade.
D
- Data ContractAn explicit agreement about a dataset's schema, semantics, quality rules and change process between producer and consumer.
- Data LineageThe record of where data came from, what transformed it, and where it flows - traceable end to end.
- Dead-Letter Queue (DLQ)A holding area for messages or jobs that failed processing after their retry budget, kept for inspection and re-driving.
- DiscoveryThe structured process of understanding a customer's current workflow, pain, constraints and success criteria before any solution design.
- Domain KnowledgeUnderstanding of the customer's industry, terminology, regulations and workflow economics - beyond the technology.
E
G
H
I
L
O
P
- Production ReadinessThe state where a system can run reliably in a real environment: evaluated, observable, rollback-tested, documented and supported.
- Proof of Concept (POC)A deliberately scoped build that tests whether an approach can work, with explicit success criteria - not a prototype of the product.
- ProvenanceThe documented origin and processing history of a specific piece of data or a generated artifact.
R
- RAG (Retrieval-Augmented Generation)An architecture where an LLM answers using documents retrieved from a corpus, reducing fabrication by grounding responses in sources.
- RerankingA second-stage retrieval step that reorders candidate documents with a more precise (often slower) model to improve top-k quality.
- Retrieval EvaluationMeasuring whether a retrieval system surfaces the right content: hit rate@k, MRR, and per-scope breakdowns against a golden query set.
- Retry BudgetThe maximum share of requests (or attempts per job) allowed to be retries, preventing retry storms from amplifying an outage.
- RollbackReturning a system to a previous known-good version, with a tested, time-bounded procedure defined before deployment.
- RunbookOperational documentation written for a stressed human at 2 a.m.: what the system does, how it fails, how to tell, and what to do.
S
- Scope CreepThe gradual expansion of a project's commitments beyond the agreed brief, usually through undocumented small additions.
- SLI (Service Level Indicator)The measured quantity an SLO is computed from, e.g. the fraction of requests answered successfully within 300ms.
- SLO (Service Level Objective)A target reliability level for a service, expressed as a measurable threshold over a window, e.g. 99.5% of requests succeed monthly.
- Solution ValidationVerifying with evidence that the proposed solution actually solves the measured problem - before full build.
- Stakeholder MappingIdentifying everyone who affects or is affected by a delivery, their influence, and what each needs from the project.
T
- TCO (Total Cost of Ownership)The full cost of a solution over its life: build, run, licenses, maintenance, support and decommissioning.
- Technical DiscoveryThe systems-focused half of discovery: architecture, data flows, interfaces, constraints and quality of the customer's environment.
- Tenant IsolationEnsuring data and operations of one customer (tenant) can never be accessed by another, enforced in queries, storage and caches.
- Tool CallingAn LLM capability where the model requests execution of registered functions (APIs, searches, actions) with structured arguments.
- TraceThe end-to-end record of one request's path through services, with timing and context, linked by a correlation ID.