AA Essentials flashcards
AA Essentials · Build recall with a focused review session.
32 cards. Read the term and the definition here. The deck at the top of this page is optional.
- Automation
- Execution of a predefined rule or process by a system with minimal human intervention, replacing repeatable manual steps rather than judgment-based decisions.
- Deterministic automation
- An automation that produces the same output for the same input every time, making behavior predictable and testable without output validation.
- Event-driven trigger
- A trigger that starts an automation when a specific change occurs, such as a new record, an updated status, or an incoming message.
- Schedule-driven trigger
- A trigger that starts an automation at fixed time intervals, such as hourly or nightly, independent of any external event occurring.
- Idempotency
- A property where running the same automation multiple times produces the same end state, which makes safe retries possible after transient failures.
- Workflow
- A defined sequence of steps with inputs, outputs, decision branches, and error paths that implements part or all of a business process.
- Human-in-the-loop
- A control pattern where a person reviews or approves an automation's action before it takes effect, used for exceptions and high-impact operations.
- Automation ROI
- Time saved per run times runs per period times labor rate, minus build cost amortized plus run and maintenance costs.
- Model
- A function with learned parameters that maps inputs to outputs, where training adjusts those parameters to minimize a loss on data.
- Supervised learning
- A learning approach that trains on labeled input-output examples so the model can predict the label for new, unseen inputs.
- Unsupervised learning
- A learning approach that finds structure, groupings, or patterns in data without any labeled target values provided during training.
- Classification versus regression
- Classification predicts a discrete label such as spam or not spam, while regression predicts a continuous numeric value such as a price.
- Train-validation-test split
- Partitioning data so training fits parameters, validation tunes choices, and the test set estimates generalization once at the end.
- Overfitting
- A condition where a model fits training data closely but generalizes poorly, shown by low training error and high validation error.
- Next-token prediction
- The core operation of a large language model, which estimates a probability distribution over the next token given the preceding context.
- Hallucination
- Confident model output that is not grounded in source data or fact, reduced but not eliminated by retrieval grounding and citations.
- Precision and recall
- Precision is true positives divided by predicted positives; recall is true positives divided by actual positives; the trade-off depends on threshold.
- Training data bias
- Systematic skew in training examples or labels that the model learns and reproduces, making data provenance and representativeness correctness concerns.
- Automation specification
- A written description of trigger, inputs, ordered steps, decision branches, outputs, and error handling that must exist before building a workflow.
- Secrets management
- Storing credentials in a vault or secrets manager with least-privilege scoping, rather than hardcoding them in scripts or workflow definitions.
- Exponential backoff
- A retry strategy that increases wait time between attempts, usually with jitter, capped at a maximum number of tries.
- Input validation
- Checking type, range, and required fields at the automation boundary so bad data fails fast instead of propagating downstream.
- Correlation ID
- A unique identifier attached to every log entry of a run, enabling reconstruction of a single execution across systems.
- Dry-run mode
- A testing mode where an automation executes its logic against sample data without performing side effects such as writes or sends.
- Step timeout
- A configured maximum duration for a single workflow step, after which the step fails or escalates instead of hanging indefinitely.
- Composable automation
- A small, single-purpose workflow that is easier to test, monitor, and reuse than one large monolithic automation.
- Automation inventory
- A maintained register of all automations with owners, purpose, systems touched, and review dates, forming the base of governance.
- Segregation of duties
- A control ensuring the builder of a high-impact automation is not the sole approver of its production deployment.
- RPA versus API integration
- UI-level bots are brittle to interface changes, while API integrations are more stable but require supported system interfaces.
- Straight-through processing rate
- The share of transactions completed end to end by automation without human intervention or manual exception handling.
- Immutable run log
- An append-only record of automation executions retained per policy, enabling traceability from business outcome back to the run.
- Manual fallback
- A documented human procedure invoked when a critical automation fails, paired with monitoring that alerts on missed runs.