Changes in version 0.1.0.9004                      

Schema v5: entity_id indexed in FTS5 (2026-05-17)

  - meta_search_fts now indexes a fifth column — entity_id — so that
    searches on entity codes (e.g. Kolada N15023, SCB TAB4822, Trafa
    measure codes like andelbefolk) actually return hits. Previously the
    index covered only title, description, search_keywords and category;
    codes that lived only in entity_id were silently unmatched. (t10011
    happened to match because Trafa stores the product code in category
    too — that path is unchanged and still works.)
  - meta_search schema is unchanged — entity_id was already a column,
    only the FTS-tabellen needed to start indexing it.
  - Migration v4→v5 is automatic on nxt_open(): DROP + CREATE the FTS
    virtual table with the new column list and repopulate from
    meta_search. Idempotent.

                     Changes in version 0.1.0.9003                      

Schema v4: category column (2026-04-17)

  - meta_search table gains a category TEXT column for subject/topic
    classification of search results.
  - Extractors now populate category:
      - Kolada: operating_area (e.g. "Utbildning")
      - SCB/pixieweb: subject_path joined with " > " (e.g. "Befolkning >
        Folkmängd")
      - Trafa: product label (e.g. "Lastbilar")
  - FTS5 index expanded to 4 columns (title, description,
    search_keywords, category).
  - nxt_search() return tibble includes category column.
  - Migration v3→v4 is automatic on nxt_open().

                     Changes in version 0.1.0.9002                      

New features

  - Schema v3: search_keywords column + 3-column FTS5 index. meta_search
    gains a search_keywords TEXT column (default '') that participates
    in the FTS5 index alongside title and description. Existing v2
    databases are migrated on first nxt_open() — the column is added via
    ALTER TABLE and the FTS5 virtual table is rebuilt to cover all three
    columns.
  - recorded_searches table. One row per (query, source) pair that has
    been run against a live source API via the "nödutgång" live-search
    flow. Used by cron to refresh entity-search-keyword associations
    before they go stale. Schema: query, source, last_run_at, ttl_days,
    last_hit_count (PK on query, source).
  - nxt_record_search(handle, query, source, entity_ids, ttl_days
    = 30L). Appends the query token to every listed entity's
    search_keywords (deduplicated), refreshes the FTS5 index, and
    UPSERTs the query into recorded_searches. Enables local search to
    find entities a server-side live search surfaced — critical for
    PX-Web where the remote search matches fields the local index
    doesn't mirror.
  - nxt_expired_searches(handle, max_age = 30). Returns recorded
    searches whose age exceeds MIN(ttl_days, max_age), ordered
    oldest-first. Cron loops over this list, re-runs each live search,
    then calls nxt_record_search() with the fresh entity ids.

                     Changes in version 0.1.0.9001                      

Bug fixes

  - normalize_kolada() and reconstruct_kolada() now accept both year
    (produced by rKolada::get_values(simplify = TRUE)) and period
    (produced by simplify = FALSE). Previously the cache path aborted
    with "missing required columns: year" whenever the user combined
    simplify = FALSE with an nxt_handle-backed cache. The period column
    is now resolved dynamically and the original name is round-tripped
    back out, so both simplify modes cache and reconstruct correctly.

                 Changes in version 0.1.0 (2026-04-21)                  

Initial CRAN release.

Data caching (phases 0–5)

  - SQLite-backed cell-level cache for the nordstat family (rKolada,
    rTrafa, pixieweb). Designed for multi-user web applications with
    concurrent reads and writes.
  - Gatekeeper schema (cells) with sidecar cell_dims for arbitrary
    metadata dimensions.
  - Query → cells junction (query_cells) preserves the drop-in closure
    semantics used by the source packages.
  - Cell-level TTL — a query inherits freshness from any other query
    that refreshed overlapping cells. Default 30 days, configurable.
  - Optional asynchronous writes via mirai, with graceful synchronous
    fallback when mirai is not installed.
  - Integration hooks in rKolada get_values(), rTrafa get_data(), and
    pixieweb get_data() — opt-in via cache = TRUE + a .sqlite
    cache_location or an nxt_handle.

Metadata caching + search (phase 6)

  - Schema v2 — queries.kind (data | metadata) + queries.payload (BLOB)
    extend the existing table via a version-guarded migration.
  - nxt_cache_handler(kind = "metadata") stores whole objects (tibbles
    or lists) as serialized BLOBs. Lossless: preserves attributes like
    pixieweb's px_api, handles rTrafa::get_structure_raw() which returns
    a list, and roundtrips nested list columns (kpi_groups.members etc.)
    byte-identically.
  - nxt_search() — FTS5-powered typeahead across cached metadata. Search
    extractors per source (R/search.R) populate a lossy meta_search
    side-index as a write side effect. Sub-millisecond matches on
    thousands of rows. Falls back to LIKE prefix match when FTS5 is
    unavailable.
  - Integration hooks in rKolada::get_metadata(),
    rTrafa::get_products(), rTrafa::get_structure_raw(), and
    pixieweb::get_tables().
  - pixieweb::table_enrich() — per-table cache granularity. Each
    enriched table is stored under its own query_hash, so overlapping
    table_enrich() calls share rows. async = TRUE launches a mirai
    background job and returns the currently-cached subset plus a
    nxt_promise attribute that apps can bridge to promises::then() for
    Shiny integration. Resume-on-crash is automatic.