@@ -1,6 +1,6 @@
 Package: cd
 Title: Climate Departure Analysis from ERA5-Land Reanalysis
-Version: 0.6.0
+Version: 0.6.1
 Date: 2026-10-06
 Authors@R: c(
     person("Allan", "Irvine", , "al@newgraphenvironment.com", role = c("aut", "cre"),
 
@@ -1,3 +1,7 @@
+# cd 0.6.1 (2026-10-06)
+
+* Monthly `tmax` and `tmin` now average **local days** (fixed UTC−8, as `cd_extract_daily()` does), not UTC days. A UTC day ran from one afternoon peak to the next, so a hot afternoon counted twice: published tmax was 0.5-0.8 °C too high (most in summer), and local-day tmin is 0.15-0.42 °C lower. The published COGs for 1950-2025 are republished; anomalies and trends move only slightly, since the shift is nearly constant over time. The annual update now lands about a month later, when the last local day of the year is available. If you set `options(cd.cache_revalidate = FALSE)`, run `cd_cache_clear()` to drop cached UTC-day copies. ([#118](https://github.com/NewGraphEnvironment/cd/pull/118))
+
 # cd 0.6.0 (2026-10-06)
 
 * Daily air temperature at points: new `cd_extract_daily(points, from, to)` returns daily mean, maximum and minimum 2 m temperature (°C) for any set of BC points from 1950, for stream-temperature models and degree-day work. Days are **local** (fixed UTC−8), not UTC, so BC's afternoon peak is not split across two days (#37's bias, which does not cancel for absolute thresholds). Each row carries the ERA5-Land cell used; a point in a sea cell moves to the nearest land cell and is flagged `cell_moved`. It reads a new daily cube on S3 (`s3://stac-era5-land/daily/`, one COG per variable and year, outside the STAC catalog), built by `scripts/backfill_edh_daily.py` and extended by the monthly update about a month after each year ends. Why a cube: the ERA5-Land store makes one point cost a 120-day × 6.4° chunk, so one station × 20 years read directly took 201 s; from the cube it is 80 s for 24 years, and 300 stations take 130 s cold, 12 s warm. The monthly `tmax`/`tmin` layers still use UTC days (#37). Producer-side, every EDH backfiller now retries a truncated or 502 chunk instead of dying on it. ([#117](https://github.com/NewGraphEnvironment/cd/pull/117))