Changes in version 0.1.0 (2026-10-21)                  

  - Discovery recognizes Query capability tokens with spaces or case
    changes, accepts trailing slashes on service and layer URLs, and
    reports missing stable item IDs as unavailable.
  - Spatial results accept WKT2-only coordinate references and use WKT
    or another identifier when a custom reference reports WKID zero.
    Schema checks detect changed WKT definitions even when reported
    WKIDs match.
  - Ordered pagination verifies big-integer sort keys without rounding.
    Public ArcGIS URL validation rejects single-label hosts.

  - Validates comparable sort keys within and across ordered pages,
    including internally requested sort fields, and rejects detected
    ordering violations. ArcGIS error objects with code 429 now respect
    valid Retry-After headers on HTTP 200 responses. Catalog health
    reports removed object-ID fields as schema drift; transient service
    failures remain transient during schema and stable-ID checks.
    Checked-catalog fallback results retain live discovery failure
    details. Source-dependent NULL placement and text collation can
    limit ordering verification.
  - Retries rate limits, temporary connection failures and timeouts,
    HTTP 500, 502, 503, and 504, ArcGIS error objects with those codes
    inside HTTP 200 responses, and generic transient ArcGIS query errors
    with bounded exponential backoff and jitter. Valid Retry-After
    delays and HTTP dates are honored within a bounded wait. Permanent
    HTTP and ArcGIS errors fail without retrying. Error messages
    identify the failed request and recovery attempts.
    options(tampaBayOpenData.max_retries = 3L) controls additional
    retries per HTTP transport or ArcGIS-body error sequence (integer 0
    through 10); existing public function signatures are unchanged.
  - Uses service-advertised record limits and pagination capabilities.
    Where ordering and result offsets are unavailable, retrieval uses
    object-ID batches. A small automatic preview can fall back to
    object-ID batches if a service advertises but ignores offsets;
    explicit ordering still errors. Truncated or unexpectedly short
    object-ID batches are split. Persistently missing records, duplicate
    or unexpected IDs, and repeated ordered pages stop with an integrity
    error. Downloads retain validated chunks for resumption and write an
    atomic complete.rds marker only after all chunks validate. Resume
    clears that marker before chunk revalidation and recreates it on
    success.
  - Includes a checked catalog of 52 ArcGIS layers from Tampa, St.
    Petersburg, Clearwater, Hillsborough County, Pinellas County, and
    the Tampa Bay Regional Planning Council. The catalog policy records
    selection criteria, source limits, candidate decisions, and topic
    coverage. The catalog is selective; it does not mirror publisher
    data.
  - Exports 16 consistently named tbod_* functions for discovery,
    inspection, retrieval, and download. Twelve unprefixed development
    names were removed before the first release. See the API migration
    guide for their replacements.
  - Searches the six configured public ArcGIS organizations as well as
    the offline checked catalog. Catalog searches support jurisdiction,
    publisher, validation status, geometry, topic, category, and
    item-date filters. Live scans default to 25 service items per
    organization and report truncation, issues, and failed portals. A
    bounded scan can miss matches.
  - Adds tbod_discover() for another public ArcGIS sharing REST root,
    FeatureServer, MapServer, or layer URL. Queryable MapServer layers
    and FeatureServer layers use the same retrieval API. Discovered rows
    retain live item dates, tags, and categories; checked matches retain
    package IDs, scope, and terms. Direct URLs have not_checked status.
  - Compares current fields, object ID, geometry, CRS, and endpoint
    metadata against saved schemas before checked retrieval and
    downloads. Added fields or changed aliases warn; removed or
    type-changed fields, geometry or CRS changes, and vanished layers
    stop retrieval. tbod_schema() and tbod_check_schema() expose the
    metadata and drift report.
  - Retrieves tibbles or optional sf results with field selection,
    ArcGIS filters, ordering, row limits, and source provenance. Count
    and object-ID checks catch missing, duplicate, or unexpected pages.
    Eligible small previews can use a subset manifest; complete
    retrieval checks the full matching manifest when records exist;
    zero-match retrieval and downloads skip the object-ID request and
    report count-only integrity. Deliberate subsets remain marked
    incomplete. Sources can still change record attributes between
    requests.
  - Parses declared ArcGIS field types, dates, geometry, and spatial
    references. Date fields with unknown time-zone semantics remain raw
    milliseconds, except UTC editor-tracking fields identified by
    metadata. Request and operation timeouts bound live work.
  - Adds resumable RDS chunk downloads. Resuming rechecks current schema
    and full record membership, verifies saved checksums and provenance,
    and skips completed chunks. Atomic writes and an exclusive directory
    lock protect progress; live record attributes may differ across
    chunks or sessions.
  - Moves Tampa parks and fire-stations to their dedicated services
    after matching source IDs. The current fire-station fields are
    unqualified (for example, NAME); older joined GIS.GovServiceInfo.*
    fields are unavailable. Terms were updated for the replacement
    endpoints. The source data is not labeled CC0 by the package.
  - Adds offline catalog and schema checks, bounded twice-weekly live
    checks across all six publishers, and a separate manual integration
    suite. The validation record gives dated evidence; live services can
    change afterward. CI actions are pinned to verified commits.
  - Adds tbod_check_catalog() for a point-in-time live metadata and
    schema report across selected checked layers. It returns one
    diagnostic row per layer, including unavailable, unsupported,
    changed-schema, and unknown outcomes, without changing the bundled
    catalog. Missing schema baselines remain reportable as unknown;
    confirmed schema drift remains visible when Query capability is
    unreported. A separate weekly workflow saves timestamped RDS, CSV,
    and Markdown reports as run artifacts.
  - Generates the Data coverage dashboard from the bundled checked
    catalog and an optional recent catalog-health report. Missing,
    stale, and unknown health results display as unverified; the site
    build makes no live ArcGIS requests. The dashboard methodology
    explains the counts, limitations, and local build command.
  - Adds three offline-build-friendly tutorials for housing and
    development, Hillsborough water-quality observations, and Tampa
    capital projects. Each uses checked catalog IDs and source fields,
    shows reproducible live analysis code and a ggplot2 visualization,
    and explains source and sampling limits.
  - Rejects malformed metadata and query responses, duplicate JSON keys,
    conflicting CRS identifiers, and invalid numeric object IDs. HTTP
    requests do not follow redirects; retry waits are capped at 30
    seconds, accepted responses at 50 MiB, and JSON nesting at 64
    containers. JSON is parsed as literal text, and spatial WKT must be
    inline. The package requires httr2 1.2.3 or later and jsonlite 1.6.0
    or later; older native curl builds still need decompression security
    updates.
  - Credits nycOpenData as design inspiration. The ArcGIS client code is
    independently implemented.