From acceptance criteria to zoom testing, 1,075 testing terms defined by engineers, not marketers. Browse by letter, filter by category, or search for exactly what you need.
Trending:Agentic TestingFlaky TestTest OracleExploratory TestingService Level Objective (SLO)Regression Testing
Last reviewed July 2026 · 21 categories
This is a curated reference of 1,075 English testing terms spanning 21 categories: test planning, design and coverage techniques, test types and levels, automation and tooling, API, web, mobile, data, security, accessibility, performance, reliability, observability, defect management, metrics, and 111 terms specific to agentic and AI testing. Definitions are original editorial syntheses informed by ISTQB, ISO/IEC 25010, W3C, OWASP, NIST, OpenTelemetry, and other primary sources listed at the bottom of this page — they are not copied normative text.
Every term has a stable anchor link. Deep-link any definition as byebugs.ai/glossary#term-slug — for example #agentic-testing.
Put the vocabulary to work
ByeBugs is an agentic AI QA platform. An AI agent tests your web app, mobile app and AI features on every pull request, explores paths nobody scripted, and reports only the bugs that reproduce — each one backed by a recording and captured evidence.
Agentic testing is goal-directed software testing in which AI agents plan, act, observe, adapt, and report findings across multi-step workflows. Instead of replaying a fixed script, the agent is given an objective and chooses its own navigation, data, and checks based on what it observes, then collects evidence for every finding it reports.
Traditional test automation is open-loop: the actions are fixed in advance and do not adapt to what happens during execution. Agentic testing is closed-loop: each observation feeds back into the next decision, so the agent can explore unscripted paths, recover from unexpected states, and discover defects that were never written into a test case.
An error is the human action or decision that produces an incorrect result. A fault is the resulting incorrect code, data, or configuration. A defect is the flaw in the work product that may cause unacceptable behavior. A failure is the observable event in which the system does not perform a required function within specified limits.
Verification asks whether a work product conforms to its specified requirements, design, rules, or other defined criteria — building the thing right. Validation asks whether the completed or evolving product fulfills its intended use and stakeholder needs in the real context — building the right thing.
An evidence-grounded finding is one whose claims are directly supported by reproducible steps, captured state, logs, screenshots, traces, or other artifacts. The evidence chain connects observed behavior through captured artifacts to the reported conclusion, so a reviewer can confirm the defect without re-running the investigation.
Agent evaluation measures task success, safety, reliability, efficiency, and behavior under defined conditions. Common approaches include end-state evaluation (scoring the final system state), trajectory evaluation (scoring the sequence of actions taken), step-level evaluation of individual tool calls, and metrics such as task success rate, pass@k, bug detection precision, and reproducibility rate.
Scope: 1,075 English terms that directly support test planning, execution, automated bug discovery, diagnosis, reporting, and triage. Excluded: named attacks, exploits, vulnerability patterns, offensive-security concepts, and generic technology terms with no direct testing or bug-workflow purpose. Definitions are original, concise editorial syntheses informed by the references below; they are not copied normative definitions.
Last reviewed July 2026.