aradR 0.2.0 (release candidate)

Author and maintainer: Simona Malovana

Discovery UX

Added arad_catalog() for one-row-per-indicator browsing within an ARAD set, base, selection or explicit indicator list.
Added arad_find() as a higher-level human-readable search across indicator names/IDs, ARAD hierarchy paths and, by default, base/dimension labels and values.
arad_find() now returns deterministic, explainable relevance ordering through relevance_score and matched_in; exact ID/name matches rank ahead of partial name/ID, path and dimension matches, with stable tie-breaking.
arad_find(details = TRUE) enriches only matched indicators with available data boundaries, latest update timestamp and snapshot-context count.
Added arad_info() for inspecting selected candidate series through a compact summary plus detailed dimensions and raw update rows.
Kept arad_search() unchanged as the fast, lightweight name/ID search for backwards compatibility.
Clarified that ARAD-reported data_to can extend into forecast/report horizons and is not necessarily the latest observed historical date.
Updated onboarding and reference documentation around a browse → find → inspect → retrieve workflow.

Scoped aradR 0.2.0 to the public Czech National Bank ARAD API only.
Removed the prerelease arad_use_internal() / arad_use_external() endpoint-mode helpers and integrated Negotiate/proxy-bypass transport so private-network authentication can evolve separately from the public package.
The default endpoint can no longer be redirected through the legacy aradR.base_url option; explicit base_url arguments remain available for testing and advanced request control.
Improved pre-response HTTP diagnostics so proxy/connection failures retain the underlying redacted error instead of collapsing to a generic message; HTTP 401 responses are classified as authentication errors with public-API-key guidance.
Added an early compatibility check for the HTTP stack: legacy httr2 0.2.2 remains supported, while httr2 >= 1.2.0 is paired with its required curl >= 6.4.0; incompatible installations now fail with an actionable update-and-restart message.

Live 0.2.0 UX acceptance covers catalogue browsing and end-to-end find → info → get → wide workflows in balance-of-payments, financial-accounts and state-budget scopes.
Added compatibility with the current /indicators-dims base_code response while preserving the stable public base_id column; legacy base_id responses remain supported.
Release-candidate CI validates package checks across Ubuntu, macOS and Windows and builds the pkgdown documentation site before release.

Author and maintainer: Simona Malovana

Added the initial public API: arad_get() and arad_updates().
Added strict exactly-one-selector validation for indicator, set, base, and user-selection requests.
Added character-first ARAD CSV ingestion followed by explicit date and numeric parsing.
Added integrity checks for required columns, malformed non-missing values, and conflicting duplicate observation keys.
Added deterministic chunking for long histories and /updates-based boundary discovery.
Added safe collapse of identical cross-chunk boundary overlaps.
Added request retries, API-key redaction, retrieval diagnostics, unit tests, and opt-in live regression tooling.

Added arad_indicators() for basic indicator metadata.
Added arad_dimensions() for base/dimension metadata.
Added arad_tree() for ARAD hierarchy paths.
Added arad_snapshots() for snapshot discovery.
Added arad_search() for scoped local search over indicator IDs and names, including frequency filtering.

Added arad_wide() for safe long-to-wide reshaping with automatic snapshot disambiguation.
Added an end-to-end Get Started vignette.
Added a troubleshooting guide, contribution guidance, package citation metadata, and pkgdown site configuration.
Reworked the README around installation and a discovery → retrieval → reshape workflow.

Added explicit none, session, and disk raw-response cache modes; caching remains disabled by default.
Added arad_cache_clear() and configurable cache age/location.
Cache entries are credential-separated using a one-way API-key hash; API keys are not persisted.
Added automated tests for metadata schemas, caching behaviour, malformed metadata, and wide reshaping.

Added stratified audit sampling and comparison logic that distinguishes key, missing-value, and numeric mismatches.
Completed the initial M/Q/Y/D, multi-indicator and snapshot-backed reliability calibration.
Retained the 3650-day production chunk size after exact agreement with finer references across the completed audit matrix.
CI is cost-controlled: routine package PRs use one Ubuntu check, while the broader live audit and multi-OS checks are reserved for justified reliability/release validation.