NEWS

---

AI4OfficialStats 0.2.0 (2026-09-24)

CRAN portability fixes

• Fixed natural-language parsing so age ranges such as 15-24 are excluded without discarding a later statistical value such as 20.5.
• Converted non-ASCII characters in R source code to portable Unicode escape sequences.
• Added source-build packaging metadata for CRAN checks.

Search-first claim verification

• Added classify_stat_claim() for deterministic extraction of domain, indicator, geography, year/latest-time intent, value, unit, and explicitly named source from natural-language statistical statements, including Persian digits and common Persian statistical terms.
• Added suggest_stat_providers() and search_claim_evidence() to route parsed claims to appropriate searchable official-statistics catalogues.
• Added verify_claim() as the high-level search-first verification interface. It can automatically search/fetch official evidence or verify against supplied local evidence. Matching uses indicator, geography, time, unit, and aggregate-dimension signals rather than numeric equality alone.
• verify_claim() now treats source attribution as optional: providers = "auto" selects suitable automatic-verification sources when none is named, while a source explicitly named in the claim is respected by default. Added providers = "all" for separate multi-source verification and source-selection metadata in printed results.
• Claims no longer need to specify a reference year. When no year or explicit latest-time phrase is present, verify_claim() uses the latest available matching official observation and records that the time choice was inferred; printed results disclose both the automatic time policy and the resolved official year.
• Added verify_text() for multi-claim paragraphs.
• Added country-name/ISO-3 resolution for 249 ISO countries and a Persian alias layer for commonly used country names.
• Added magnitude-aware parsing (thousand, million, billion, trillion) and precision-aware numeric tolerances.
• Existing 0.1.0 constructors, audits, and provider connectors remain available alongside the new Easy Mode and generic-provider workflows.
• Added Eurostat Catalogue API discovery so explicitly attributed Eurostat claims can be searched and fetched automatically through verify_claim().
• official_stat_providers() now distinguishes direct fetch, catalogue search, and full automatic-verification capability instead of implying that every listed provider supports the same workflow.

Easier first-use workflow

• Adds audit_stats() as a one-call beginner-facing wrapper over the existing explicit stat_reference(), stat_claim(), and audit_stat_ai() workflow.
• Adds read_official_stats() for ordinary data frames, local files, and HTTP(S) URLs without requiring the package's canonical provider schema in advance.
• Adds detect_stat_schema() with automatic recognition of common English and Persian statistical field names.
• Adds Persian/Arabic digit normalization and common Persian unit recognition, including percent and person-count labels.
• Adds as_stat_claim() for structured rows, named lists, numeric values, and simple plain-text statistical claims.
• Extends as_stat_reference() so ordinary data frames can be converted using automatic or explicit mappings, with clearer missing-field diagnostics.

Universal provider layer

• Adds register_official_provider() for session-local generic provider profiles. Profiles can store reusable column mappings and conservative host patterns without requiring a dedicated package release for every NSO, ministry, central bank, or official organisation.
• Adds detect_stat_provider() and generic file/URL importing for arbitrary official providers.
• official_stat_providers() now advertises generic-import support and includes profiles for the Statistical Center of Iran, Central Bank of the Islamic Republic of Iran, and Iran Ministry of Health and Medical Education.
• Excel import is available through the suggested readxl package; CSV, TSV, TXT, JSON, RDS, and supported ZIP contents use existing/base dependencies.

WHO support

• Adds who_odata_url(), search_who(), and fetch_who() for World Health Organization indicator metadata and bounded OData retrieval.
• Adds WHO to search_official_stats() and the built-in provider registry.
• The WHO OData base URL is configurable through options(AI4OfficialStats.who_base_url = ...) or base_url, allowing users to move from the legacy GHO OData endpoint to the World Health Data Hub OData implementation as WHO completes its API transition.

Compatibility

• Existing 0.1.0 constructors, audits, proof bundles, provider connectors, benchmark tools, and GSBPM alignment functions remain available.
• New convenience functions are additive; the explicit advanced API remains the reproducibility-oriented path for users who want complete control.

AI4OfficialStats 0.1.0 (2026-09-22)

Release scope

• Provides a ten-dimensional statistical-fidelity audit covering semantic, numerical, temporal, geographic, unit, contextual, provenance, revision, transformation, and uncertainty fidelity.
• Provides geometric and weakest-link summaries together with non-compensatory critical-error rules and PASS/WARN/FAIL certification logic.
• Provides provenance graphs, revision handling, portable SHA-256 proof bundles, batch auditing, prompt perturbation, statistical red-team generation, minimal-pair tests, benchmark utilities, and model-stability tools.
• Provides bounded interfaces for Eurostat, World Bank, OECD, UNSD SDG, UK ONS, U.S. BLS, and low-level IMF SDMX URL construction, together with an extensible HTTPS JSON API registry.
• Provides gsbpm52_map(), gsbpm_align(), and gsbpm_quality_report() for explicit alignment with relevant GSBPM 5.2 Analyse, Disseminate, Evaluate, Quality Management, and Metadata Management activities.
• Includes offline fixtures and regression tests for provider parsing, audit behaviour, proof integrity, and GSBPM alignment.

Scope and compatibility

• GSBPM alignment is offline and deterministic and introduces no additional network dependency.
• The package supports relevant GSBPM 5.2 activities but does not claim to implement or certify the complete GSBPM statistical production process.