Check hex7 on sample and sample_root against the positions and against the observations

The release gate for add_sample_hex7(). One row per rule and table:

Usage

check_sample_hex7(
  con,
  sample_tbl = "sample",
  root_tbl = "sample_root",
  obs_tbls = c("obs_bio", "obs_env")
)

Arguments

con

DuckDB connection.

sample_tbl, root_tbl

the stamped sample and the build_sample_root() table.

obs_tbls

observation tables carrying sample_key, latitude, longitude, hex7; those absent from con are skipped.

Value

A tibble check, table, n, n_bad, status (ok | fail | report). Assert all(status != "fail").

Details

• position_has_hex7 — n rows with a finite position, n_bad of them without a hex7;

• hex7_has_position — n rows with a hex7, n_bad of them without a finite position. Together these two say count(hex7) equals the count of finite positions, row for row;

• hex7_is_res7 — n rows with a hex7, n_bad whose H3 resolution field is not 7;

• root_equals_sample — n root samples (no parent), n_bad roots whose hex7 differs between the two tables, plus roots missing from root_tbl, plus root_tbl rows that are not a root sample;

• obs_same_position (per table of obs_tbls present) — n observations sitting at exactly their own sample's position, n_bad of them in another cell than their sample. This is the proof, on the data, that the sample side and the observation side share one definition;

• obs_other_cell — n observations where both the row and its sample carry a cell, n_bad in another cell than their sample. Reported, never a failure: an observation carries its own position (each scan of a CTD cast, a DIC draw), the sample carries the event's, and a cast that drifts across a cell edge puts some of its scans in the neighbouring hexagon.

A finite position is both coordinates present and neither NaN nor infinite. Needs no extension: the resolution is read from the cell's own bits.