Glossary
Agent. An AI system that calls tools to complete a task; here an assistant that calls the organization's data API or MCP server to answer a question about statistics.
AI-ready. Data and metadata that software, including AI systems, can find, retrieve, interpret, and cite without manual steps. See the AI-ready data framework.
Annotation. Optional properties on a tool that describe its behaviour (for example that it is read-only); hints that clients treat as untrusted unless the server is trusted.
API. Application programming interface: the address at which a program asks a system for data and receives it in a structured form.
Artifact class. The kind of object a schema describes, defined by its analytical purpose, boundaries, and expected diversity. A member of the class is an artifact.
Calibration. The comparison of model scores with curator scores on a sample, per dimension, to find systematic bias.
Candidate. A model shortlisted for a task, with its licence classified and its card checked, to be run on the suite.
Checksum. A fingerprint of a file's content (here SHA-256) that changes if one byte of the file changes; used to prove that a file is as released.
Concept coverage. The descriptive evidence obtained by assigning development field profiles to reviewed concepts or a residual category; it supports human review.
Concept inventory. An intermediate collection of descriptive concepts, definitions, examples, and refinement decisions that precedes operational field design.
Confidence. A number between 0 and 1 that a model returns with its output to say how sure it is; the basis of a threshold rule.
Coverage. The share of a reference list of known uses that the collected documents contain.
Crosswalk. A mapping from one classification to another, for example from a national occupation classification to ISCO.
Data snapshot. A bounded visual region of a document page containing structured or semi-structured information for analytical interpretation or operational reuse, such as a statistical table, chart, analytical map, or monitoring dashboard. A crop should retain the context needed to interpret it.
DDI. The Data Documentation Initiative, the standard for documenting surveys and their variables, value labels, and files; the World Bank microdata schema is based on it.
Decision. A curator's recorded verdict on a suggestion: accept, edit (with the final text), or reject (with a reason).
Document record. The catalog record of a document in the World Bank document schema, with the series and surveys it draws on.
Estimated. The status of an extracted value for which the document offers no check: no printed label, total, or derivation.
Evidence pack. The list, by assessment question, of the artefact that answers it, its location, and its date, compiled for an AI-readiness assessment.
Flagged. The status of an extracted value that failed a check and awaits correction with the page open.
Gap. The distance between a dimension's current level and its target, measured in levels and in the number of questions below the target's value.
Gate. The rule that decides whether a change ships, applied to a paired comparison of the evaluation suite before and after the change.
Grounding. Restricting a language model to what was retrieved or given for the task: an answer that uses only the retrieved records, a draft that uses only what the record and its named sources support.
Held-out validation. Assessment of an operational schema against artifacts excluded from its initial development, with candidate findings and explicit claim limits.
Human adjudication. A documented human decision about the evidence, scope, materiality, and appropriate treatment of a proposed concept or schema change.
Incident. An AI component's output reaching a reviewer, a product, or a user with an error that the controls should have caught, or that a control caught at the last step; handled by a fixed procedure and recorded.
Inventory. The list of documents with their type, year, page count, text-layer status, and counts of tables and figures.
Issue. The review pipeline's output: a detected problem with a category, a severity from 1 to 5, the current value, and a proposed value.
Label. The column a model is trained to predict: the occupation code, the transcribed value, the series a question refers to.
Language model. A model that reads and writes text; used in these guides to draft, flag, code, explain, and translate, with a person deciding.
Layout detection. A model that finds regions of a page image and returns their class and bounding box.
LLM. Large language model. A model that generates text, used here for drafting metadata, answering questions, and grading.
Manifest. A file that lists the parts of something so that it can be checked or resumed: for a job, the records and their status; for a release, every file with its checksum; for an agent interface, the design document of its tools, inputs, outputs, examples, and resources.
Map. The table of model-assisted tasks with phase, role, review, data sensitivity, model location, and owner.
Maturity level. One of three states used in this guide: foundational, AI-ready, AI-native.
Metadata observation. An artifact-level proposal for a reusable descriptive property, retaining its observed value, evidence source, description, and rationale.
ML-ready dataset. A fixed, versioned set of examples with a documented structure, a stated purpose, a split for training and testing, a statement of whom the examples represent, and a licence that covers model training.
Operational metadata schema. A specification of usable metadata fields, definitions, evidence rules, examples, and annotation guidance for an intended application.
Precision. The share of extracted mentions that are correct.
Precision and recall. Precision is the share of what a system found that is right; recall is the share of what is there that the system found. F1 is the average that combines the two.
Prompt. The instructions and context given to a language model for one task; kept as a versioned file when it is part of a workflow.
Provenance. The record of where a value came from: the series, the release, the source, the document and page, the method, and the version, carried with the value.
RAG. Retrieval-augmented generation. A pattern where the system retrieves records first and the model writes the answer from them.
Recall. The share of labelled mentions that the extractor found.
Representativeness. The degree to which a dataset's composition matches the population it will be used on, measured group by group and stated in the card.
Resource. Context a server offers to the model or the user, such as code lists or a release calendar, identified by a URI.
Roles. What a model may do in production: flag, propose, code, explain, draft. It does not decide.
schema.org. A vocabulary that search engines agree on for describing things on web pages; a Dataset record in it, embedded in a page, is how dataset search engines learn what the page is about.
Split. The division of a dataset into a training part (the model learns from it), a validation part (the model builder chooses settings on it), and a test part (touched once, to report a score).
Structured output. A model's answer returned as data in a declared shape (a JSON object with named fields that conforms to a schema) so that a script can validate, read, and aggregate it.
Study record. The study-level documentation of a survey: title, abstract, dates, coverage, universe, design, access conditions.
Suite. A question set, the scripts that score it, the required scores, and the record of runs.
Table record. The catalog record of a statistical table in the World Bank table schema, with columns, rows, sources, definitions, citation, and relations.
Threshold. The value of a score (a confidence, a similarity, a garble rate) at which the rule changes what happens to an item; set from measurements.
Token. The unit in which language models read and write text, roughly three quarters of a word; prices, context limits, and throughput are counted in tokens.
Trace. The logged record of one question: tools called, series used, answer, whether the agent declined, latency.
Use. A document whose analysis, decision, or statement depends on the data, as opposed to a mention that cites them as context.
Utility. How closely the synthetic file reproduces the real one at the marginal, association, and analysis levels.
Variant. A way of naming a dataset other than its canonical name: an acronym, an informal name, a translation, a misspelling.
Vocabulary. A concept scheme with preferred labels, alternates, and URIs, to which free-text keywords and topics are mapped.