• Fix: Honour every percentile requested from generic percentile while preserving the 7.3.1 single-percentile output shape (#557). From at least 7.2.0 through 7.3.1, the source used only the first requested value (7.2.0 source read).
• Fix: Average the ECA&D leave-one-year-out bootstrap index estimates for temperature percentile counts TG10p/TG90p, TN10p/TN90p, and TX10p/TX90p (#540, #545). This is a 7.3.0 regression: running 7.2.0 gives the correct ATBD value, while 7.3.0 and 7.3.1 are affected.
• Fix: Restrict the precipitation operand of CD and WD to wet days (RR >= 1 mm) (#541, #546). Affected in 7.2.0 and 7.3.1 (run).
• Fix: Apply the requested percentile base period, interpolation, day-of-year window and only_leap_years setting to the default compound thresholds used by CD, CW, WD and WW (#560, #546). Affected in 7.2.0 and 7.3.1 (run).
• Fix: Use the strict ECA&D TG > 5 degC GSL onset and honor disabled completeness for isolated missing values (#537, #538, #548). Affected since at least 7.2.0 through 7.3.1; the 7.2.0 source was read, not run.
• Fix: Accept CF precipitation-flux units in RX5day (#544, #549). Verified by running 7.3.1; the 7.2.0 source was read, but no earlier range is claimed.
• Known issue: CD, WD and WW can differ by one day at the edges of a subset percentile base period because the five-day calendar window is truncated instead of wrapped (#559).