v0.6.13

• The README is rewritten in plain language: what cd measures and the restoration decisions it informs, daily point temperatures and the baselines for any annual series, a figure and an example with a 1951–1980 baseline built from one committed script (data-raw/readme_figure.R). The DESCRIPTION Title and Description, the README tagline and the GitHub About now say the same thing. A new NOTICE.md attributes ERA5-Land (CC BY 4.0) and the BC Freshwater Atlas and ecoregion data (Open Government Licence – British Columbia) that the bundled data are built from, and the pkgdown reference index is grouped by purpose. (#142, #152)
• citation("cd") uses the key cd and the note "Version x", so knitr::write_bib() keeps the entry and a report can cite @cd.

v0.6.12

• Rewrites the sentences in the Kootenay Lake article that its own data contradicts. Soil moisture is declining, not flat. The warming range comes from the three ecoregions shown. The watershed groups agree in direction but only KOTL is significant. Both articles introduce four findings. The prose counts the three ecoregions it draws. The seasonal snow caption no longer carries the Peace numbers, and the Kootenay ecoregion filter moves to the Peace article's 100 km², which draws the same three ecoregions. (#146, #149)

v0.6.11

• The two regional articles' per-ecoregion roll-up tables labelled the precipitation trend as mm per year; it is a trend of percent anomalies, so the column is now prcp %/yr and the caption reads "percent of the 1951-1980 baseline mean per year". The numbers are unchanged. The Kootenay article's precipitation captions now describe its own ecoregions: the facet caption no longer names the Peace ecoregions BMP and NRM, PTR is Purcell Transitional Ranges on the eastern edge, and the captions cover the three ecoregions drawn (Thompson-Okanagan Plateau falls under the 50 km² filter). (#102, #147)

v0.6.10

• cd_variables() gives swe_max (annual peak SWE) the anomaly type pct_normal and unit "%", as swe already had, so cd_anomaly() and cd_trend() report peak snowpack as percent of the baseline normal rather than mm, clamped at cap_pct (±200 by default) like the other pct_normal variables. On raw swe_max values, cd_summary() and cd_plot_timeseries() no longer show a unit, the same as for swe. A table saved under cd 0.6.9 or earlier keeps the absolute/"mm" it carries, because a carried type wins over the registry, and binding it to new output now errors (two types in one series); recompute it from the extracted series. The two regional articles' peak-SWE figures are now in percent, and their other numbers are unchanged. (#139, #145)
• cd_compare()'s docs now call its default window_a = 2015:2025 the recent 11 years rather than a decade. Behaviour is unchanged.

v0.6.9

• cd_compare(method = "pct_change") now returns difference = NA, with one warning naming the rows, for a variable whose values have no real zero or are already in percent: tmean, tmax, tmin, rh, snow_cover, snowfall_fraction and snowmelt_doy_50. A temperature's percent depends on where 0 °C falls, so a base mean of −0.57 °C read as +343 %. Callers that asked for a temperature percent now get NA and the warning; use method = "mean_diff" for those. The rule is a new logical column, cd_variables()$pct_change, not the anomaly type: vpd, swe_max and snowmelt_rate_peak are absolute and still take a percent. A series from outside the registry can carry its own pct_change column (see the input contract in ?cd_anomaly); one measured in °C must carry pct_change = FALSE. (#138, #143)

v0.6.8

• New article, How the Weather Has Changed in Three Northern BC Watersheds: a short, plain-language story of the Bulkley, Nechako and Parsnip watershed groups against 1951–1980, written for readers with no climate background, with the upper and lower Bulkley compared and a table of what the changes mean for streams and fish. Every number on the page is computed from bundled data, and the data script stops a regen that would make one of its sentences false. (#137, #141)
• cd_compare() now describes its default reference window, 1951–1980, as the pre-warming reference (the GISS base period) rather than a WMO-style standard normal. Behaviour is unchanged.

v0.6.7

• cd_variables() gives vpd the unit "hPa", the unit its data has always been in. It said "Pa", so the vpd Unit from cd_anomaly(), cd_trend(), cd_summary() and on the cd_plot_timeseries() axis was off by 100×. Values are unchanged. A table saved under an older cd keeps the "Pa" it carries, because a carried unit wins over the registry, and binding it to new output now errors (two units in one series). Relabel it with x$unit[x$variable == "vpd"] <- "hPa", or recompute from the extracted series. (#96, #134)

v0.6.6

• cd_cache_fetch() (so every cache = TRUE extract) and cd_stac_catalog() now work on R 4.1-4.3, as Depends: R (>= 4.1) promises: %||% is now imported from rlang, since base R exports it only from 4.4.0. (#130, #133)

v0.6.5

• cd_catalog() gains checksum, size, cd_version, cd_sha, run_time and run_url, and cd_extract() and cd_extract_daily() gain checksum and cd_sha on every row, so an analysis can record which build of each file it read and notice a republish. cd_sha is read from the file itself; the checksum is kept only when the file is shown to be the build the catalog or manifest describes, so a stale cached copy gives NA and a warning rather than another build's checksum. A checksum identifies a build, not values. The vignettes' precomputed series lack the new columns until data-raw/*_vignette_data.R is re-run. (#126, #129)
• cd_extract() with no matching catalog rows now returns a typed zero-row tibble rather than a 0 x 0 one. The tests need testthat (>= 3.2.0).

v0.6.4

• Run provenance in catalog.json and daily/manifest.json moves to the STAC processing extension (v1.2.0), a standard any STAC reader understands, in place of the four cd: fields added in 0.6.3. The fields are now processing:version, processing:software ({"cd": "<sha>"}) and processing:datetime (the COG's CD_RUN_TIME tag, copied exactly), plus a processing-execution link to the GitHub Actions run that built the file. cd:variable and cd:period are unchanged. The COGs are not rebuilt, so their checksums and ETags stay the same and no cache downloads them again; only the two metadata documents are republished. (#128, #131)
• Producer side, the catalog and the manifest are both refused unless every item or entry carries the new keys, so a partial publish cannot mix the two vocabularies. scripts/catalog_publish.R and scripts/daily_publish.R --manifest-only republish the metadata alone and never upload a COG.