Glossary

Definitions for the terms you’ll meet across the aha pipeline.

Key terms

Codebook — Data Dictionary: A structured document describing the variables in a dataset — names, definitions, data types, coded values, and special codes for missing or sentinel values. aha uses codebook information to understand the datasets and inform the plan.
Data type: The format in which a variable’s values are stored. Common types include categorical_nominal, binary, numeric, text, and datetime.
Decision type: One of five categories assigned to each variable: Matched, Mapped, Partially Mapped, Calculated, or Not Mapped. Describes how the source variable relates to the standard.
Harmonization: The process of transforming a source dataset so its variables, values, and structure conform to a defined standard schema.
Harmonization rate: The percentage of source variables that passed all five validation checks after execution.
Human intervention: A flag indicating that a variable’s plan confidence is below the researcher-defined threshold and requires manual review before proceeding.
Missing value code: A numeric or symbolic code used in the dataset to represent a missing or unknown value (e.g. -4, 88, 99).
Plan confidence: The confidence that a plan is correct.
Sentinel value: A specific code used to indicate a special condition — such as not applicable or not assessed — rather than a true missing value.
Standard dataset: The reference schema defining the expected variable names, data types, and coded values that the harmonized output must conform to.
Threshold: The plan confidence cutoff below which variables are flagged for human review. Set in the Strategy & Planning module.
Transformation: A sequence of operations applied to a source variable to convert it into the format required by the standard target.
Unharmonized dataset: The raw, study-specific source dataset uploaded by the researcher before any harmonization has been applied.
Validation check: One of five criteria evaluated in the Validation module: concept alignment, data type compatibility, value/category coverage, missing & placeholders, and stats sanity check.