Changes in version 0.1.0                        

  - Initial version following the project plan.
  - Fetch layer: fetch_eia(), fetch_bea(), fetch_census() and
    fetch_rps(), with a local download cache and graceful failure when a
    service is unavailable.
  - Harmonise layer: harmonise_states(), deflate() and convert_units().
  - Panel builder: build_panel().
  - RPS stringency module: rps_rules() and rps_stringency() with the
    adoption, nominal, coverage_adjusted, incremental and flexibility
    methods.
  - Documentation helpers codebook() and cite_sources(); output helpers
    as_pdata() and export_panel().
  - The bundled RPS rule coding (inst/extdata/) covers all 50 states and
    DC. The 33 RPS states and 6 voluntary-goal states are verified
    against statutes, DSIRE and LBNL data (column verified), coverage
    shares are filled in, and repeals are coded (AZ 2026, KS 2015,
    MT 2021, OH 2027, TX 2025, WV 2015).
  - Tested against the live EIA and BEA APIs: BEA GDP tables now use
    their current names (SAGDP9, SAGDP2), personal income is labelled in
    millions of dollars, and requests retry on transient HTTP 500 errors
    from the EIA gateway.
  - README now shows real output from a live run, with figures in
    man/figures/.
  - fetch_census() now requires a Census API key (the API rejects
    keyless requests) and was tested live against ACS 1-year data;
    output is sorted by state and year, and unit is only set for the
    default population variable.
  - API responses that are web pages instead of data (for example a
    rejected key) now raise a clear rpspanel_http_error instead of a
    parse failure.
  - API keys are only required when data actually need downloading, so a
    cached panel can be rebuilt offline without keys.
  - Parquet export tested with arrow (values and column types round-trip
    unchanged).
  - The 12 original no-RPS rows (AL, AK, AR, FL, GA, ID, KY, LA, MS, NE,
    TN, WY) are now verified against the LBNL June 2026 data and cite
    their sources, so all 51 rows are verified. Nebraska's
    utility-adopted clean energy commitments (counted by LBNL as a CES
    from 2040) are documented but not coded as a state RPS.
  - rps_stringency() converts capacity-based (MW) standards (Texas,
    Iowa) into a share of covered sales with new sales and
    capacity_factor arguments, following Berkeley Lab's approach;
    build_panel() supplies retail sales automatically when an EIA key is
    available, so rps_nominal is no longer blank for these states.
  - New optional rules column baseline_includes_hydro: the incremental
    baseline now includes hydro for New York, whose targets count
    existing hydro (2004 baseline about 19%, matching the 19.3% legal
    baseline).
  - Fixed: a single target in force from enactment (Iowa) no longer
    breaks interpolation.
  - Texas is recoded so only the 5,880 MW (2015) schedule is binding;
    the 10,000 MW by 2025 figure is a non-binding goal (as in Berkeley
    Lab's data). Converted Texas shares now match Berkeley Lab's within
    about 0.6 percentage points in every year.