What the hub covers
The areas are listed in the order in which the work is usually done. The assessment identifies the gaps. Documentation, access, and verification address them. The cookbooks and the reference implementation give the steps. The adoption area describes the installation test, the support boundaries, and the release practice of each repository.
Assess
Readiness framework and diagnosticsDocument
Metadata standards, the Editor, and AI-assisted curationConnect
MCP and interoperable data accessVerify
Proof-Carrying Numbers and checked outputsImplement
Cookbooks and the reference implementationAdopt
Installation tests, support boundaries, and contributionReadiness framework and diagnostics
The assessment framework, its operational companion, and the cookbook that turns a readiness profile into an improvement pathway.
Where to start
Score the organization on the twelve dimensions with the assessment framework, or take the five-minute cookbook self-assessment for a first orientation. Read the operational companion for the dimensions with the largest gaps.
Tools
Scores a national statistical organization on two pillars and twelve dimensions, with 69 questions and a gateway question, and places it at one of four maturity levels per dimension.
For each dimension of the assessment, the workstreams, tools, and cookbook recipes that support progress to a higher level, and the evidence each produces for the assessment form.
Guides
- AI-readiness assessment frameworkTwo pillars, twelve dimensions, four maturity levels; the pathway from a score to a plan.
- Operational companionResources and evidence per dimension; effort and roles per recipe.
- AI-ready data frameworkThe attributes of AI-ready data under FAIR, and how the workstreams map to them.
Metadata standards, the Editor, and AI-assisted curation
Documentation of indicators, microdata, geospatial data, and documents in the World Bank schemas, and AI-assisted review and enrichment under curator review.
Where to start
Document each product in the World Bank schema for its type with the Metadata Editor or the CSV template, run the completeness check at the foundational level, and publish to a catalog. Add AI-assisted review and enrichment once the records exist.
Tools
- Metadata EditorDeployable
Web application for documenting microdata (DDI Codebook 2.5), indicators, documents, geospatial data, images, videos, and scripts in the World Bank metadata schemas, with validation, templates, a REST API, and publication to NADA catalogs.
- NADA catalogDeployable
Open-source data cataloguing system for surveys, indicators, documents, tables, and geospatial data, with search, managed access requests, and a catalog administration REST API with API-key authentication.
- World Bank metadata schemasDeployable
JSON Schema definitions for indicators, indicator databases, microdata (DDI), geospatial (ISO 19115), documents, tables, images, videos, scripts, and resources, with a Python library of pydantic models and Excel templates.
- Metadata ReviewerPre-release
Five-agent pipeline that scans metadata records for typos, inconsistencies, missing fields, and incorrect content, and returns ranked issues with suggested corrections for a curator to accept or reject. It runs as a Python package, through the review app, and inside the Metadata Editor as an optional add-on.
- Metadata AugmentationPre-release
Generates DDI-style variable groups for a data dictionary: embeds variable labels, clusters them, asks a model to curate each group, and checks the result for self-consistency before a curator reviews it.
Scores each metadata record on completeness, semantic alignment, specificity, and consistency with structured model output, and aggregates the scores to find systematic gaps in a catalog.
- Data SnapshotsMethod
Layout detection models locate figures and tables in PDF documents so that the data they hold can be extracted and documented; the paper benchmarks open-source models on the task.
Guides
- Metadata ReviewerFive-agent review pipeline with a user manual; available as a package, a review app, and an add-on in the Metadata Editor.
- Metadata AugmentationDDI-style variable groups generated for a data dictionary.
- Generative AI for metadata qualityScoring records on completeness, alignment, specificity, and consistency.
- Mapping to the GSBPMWhere each workstream applies in the statistical production process.
Cookbooks
- Metadata Curation with Language Models8 chapters
- Extracting Data from Documents9 chapters
- AI-Ready Data Dissemination9 chapters
MCP and interoperable data access
Access to official statistics by software and AI agents: files at stable addresses, APIs, and Model Context Protocol servers, with provenance in each response.
Where to start
Publish tidy files at stable addresses with a licence, then an API with provenance in every response, then an MCP server over the API. Connect a client to the Data360 MCP server to see the pattern before building one.
Tools
- Data360 MCP serverOperation
Model Context Protocol server over the World Bank Data360 platform: search of indicators with country coverage checks, metadata, disaggregation, codelist lookup, and time-series data for any MCP client, with chart specifications and a system-prompt resource.
- NADA AIPre-release
Semantic search and a Model Context Protocol server over a NADA catalog: the service ingests the catalog records through the NADA REST API, indexes them with local or hosted embeddings, answers keyword, vector, and hybrid queries, and exposes the catalog to MCP clients through tools, interactive apps, resources, and prompts. It is being integrated into NADA, and its release is tied to the NADA release.
- Data AI ChatbotDeployable
Self-hostable chat application that connects to any MCP server over HTTP or SSE, routes model calls through LiteLLM, renders charts and documents in a side panel, and marks numbers in answers with Proof-Carrying Numbers. The World Bank customization, Data360 Chat, ships pre-configured for the Data360 MCP server.
Architecture and tool design for an MCP server over official statistics, with a minimal example server in the cookbook and a full treatment of tools, context, access control, safety, and evaluation in the agents cookbook.
- Semantic search and the PI-FT toolkitPre-release
Indexing and search by meaning over catalog records, and a toolkit that fine-tunes embedding models on structured metadata with permutation-invariant training, reporting Recall@k, MRR, and nDCG.
Guides
- Model Context Protocol for statisticsArchitecture, tool design, and the servers in operation.
- MCP for official statisticsUse cases for statistical offices and what a server over a catalog returns.
- Data discoverabilitySearch by meaning over catalog records, and fine-tuning embeddings on structured metadata.
- PI-FT toolkitPipeline, configuration, and deployment of fine-tuned embedding models.
Cookbooks
- AI-Ready Data Dissemination9 chapters
- Serving Official Statistics to AI Agents9 chapters
- Publishing ML-Ready Datasets8 chapters
- Synthetic Data for Sharing7 chapters
Proof-Carrying Numbers and checked outputs
Verification of the numbers in generated text against the records they cite, evaluation of retrieval and answers, and quality checks on data and metadata.
Where to start
Put a source and a citation on every published figure, return provenance from the API, and verify the numbers in generated answers against the records they cite. Build an evaluation suite before a model-assisted step goes into production.
Tools
- Proof-Carrying NumbersDeployable
Presentation-layer protocol for numeric fidelity: a model emits each number as a claim tag naming its source record, and the renderer verifies the number against the record under a declared policy and draws the mark. TypeScript packages for the core, the UI, and a Data360 preset; a Python core for comparison; a specification with shared fixtures.
- Anomaly detection and explanationPre-release
Takes anomaly scores from any detector on indicator time series, builds the context for each flagged point, elicits a structured explanation from a model with a classification, a confidence, and cited evidence, and exports it to a web application where a reviewer approves, rejects, or corrects it.
- Monitoring of data usePre-release
Extracts dataset mentions from documents with zero-shot named-entity recognition, harmonizes name variants to canonical datasets, and deduplicates them, so that use of a dataset in research and policy documents can be counted.
Guides
- Proof-Carrying NumbersThe protocol, the packages, the policies, and what a verified mark does and does not mean.
- Anomaly detection and explanationEvidence-backed explanations of flagged values with a reviewer workflow.
- Monitoring of data useDataset mentions extracted from documents and harmonized.
Cookbooks
- Evaluation Suites for Statistical AI9 chapters
- Monitoring Data Use9 chapters
- Language Models in Statistical Production7 chapters
- Small and Open Models for Statistical Offices7 chapters
Cookbooks and the reference implementation
The cookbooks with their tested scripts, and the reference implementation that runs documentation, validation, publication, agent access, and verification in one pipeline.
Where to start
Run the end-to-end reference implementation on the example organization to see every step and its output. Then follow the cookbook for the data type or the service the organization is working on, with its scripts on the organization’s own files.
Tools
- End-to-end reference implementationDeployable
A script that documents a set of series and a survey in the World Bank schemas, validates them against the pinned schemas and a completeness profile, publishes a catalog with schema.org records, serves it through an MCP server, composes an answer from the tool results, and verifies the numbers in the answer with the PCN engine.
- Cookbook scriptsDeployable
The scripts shipped with the cookbooks: checkers, scorers, builders, and converters that run on the shared example organization, each with a docstring stating its purpose, usage, exit status, and what it does not check.
Guides
- End-to-end reference implementationDocumentation, validation, publication, MCP access, and verification in one run.
- Cookbook landing pageAll cookbooks, by group and by reading path.
- How the cookbooks are writtenThe standard every cookbook follows and the checks that enforce it.
Cookbooks
- AI-Ready Data Dissemination9 chapters
- Serving Official Statistics to AI Agents9 chapters
- Evaluation Suites for Statistical AI9 chapters
- Language Models in Statistical Production7 chapters
Installation tests, support boundaries, and contribution
The installation test, the support boundaries, the release practice of each repository, and where issues and contributions go.
Where to start
Read the status, the installation guide, and the known limits of a tool on the tools page. Install it from the documentation alone, record the blockers, and file the record as an issue in the tool’s repository.
The installation test
A team outside the program installs a tool from its documentation and records the result as an issue. The steps are the same for every tool.
- Choose the tool and read its row on the tools page: status, stack, licence, installation guide, and known limits.
- Install from the installation guide alone, on a machine the original developers have not touched, and note every point where the guide was not enough.
- Run the tool’s own smoke test: the test suite, the quick start, or the first recipe of the matching cookbook.
- Record the elapsed time, the operating system and versions, the blockers, and the workarounds.
- File the record as an issue in the tool’s repository with the title “Installation test: <tool> on <platform>”. A test that passed is as useful as one that failed.
Support boundaries
The program offers
- Public repositories with an issue tracker for each tool, read by its maintainers.
- Documentation on this site and in each repository: user guides, installation guides, API descriptions where the tool has an API, and the cookbooks.
- Versioned releases where the tool has them, with a documented way to pin and roll back.
- Contact with the Development Data Group for organizations planning an adoption or a joint implementation.
The program does not offer
- Hosting or operation of a tool on behalf of another organization. The services in operation serve World Bank data.
- Service levels, on-call support, or guaranteed response times on issues.
- Adaptation of a tool to an organization’s systems, languages, and classifications, which the cookbooks describe and the organization carries out, alone or in a joint engagement.
Releases and versioning by repository
| Repository | Practice |
|---|---|
| data360-mcp | Semantic versioning with git tags and GitHub releases; the Python package on PyPI; rollback by installing a prior release. |
| metadata-editor | Tagged releases with notes; upgrade by replacing the application files and running the installer. |
| metadata-schemas | Tagged releases of the Python package; the schema files on the main branch are the reference, so pin a commit for validation that must match a catalog. |
| pcn | Changesets in the monorepo; the TypeScript packages are published to npm under the @pcn-js scope with semantic versions; the Python core is installed from the repository at a pinned commit. |
| data-ai-chatbot | Releases from the repository; Docker Compose files for development and production; the deployment guide covers upgrades. |
| ai4data | The Python package on PyPI with extras per workstream; the cookbook scripts and the reference implementation are tested in CI on every change. |
Guides
- Tools catalogueEvery tool with its status, stack, licence, operator, installation guide, and known limits.
- ContributingHow changes to the methods, scripts, and documentation are proposed.
- Code of conductExpected behaviour in the repositories and their discussions.
- IssuesBug reports, installation test records, and feature requests for this repository.
Organizations planning an adoption or a joint implementation can write to the Development Data Group at aifordata@worldbank.org. Reports of installation tests and bugs go to the issue tracker of the tool’s repository, where the maintainers read them.