The published thresholds table, from a tuning run's outputs

One row per species × stage × component of the joint pick, in the columns link's calibration reads (link#346): rule_id ties the components of one rule; link_parameter is the column of link's parameters_habitat_thresholds.csv the value fills: since link 0.67.0 (fresh 0.42.0) the near-a-river exception fills <stage>_river_discharge_min and <stage>_river_distance_max, with river_measure = discharge saying which of fresh's three measures of a river the value is in (link converts to width where a group sizes streams by width); model says whether the size test is a width (cw) or a MAD (mad) floor; range_lo and range_hi are the smallest and largest value of the component among the rules inside the one-SE region on the main reading; readings_in is the number of fold seeds (the main one and the extra seeds) in whose region the recommended rule sits, of readings; rule_196 is the rule the same pick makes from a region 1.96 standard errors wide (knw_tune_repick()), and holds_196 is TRUE where that rule keeps the row's value (a size row also needs the same size variable; a component the recommended rule lacks holds when the 1.96 rule lacks it too). rule_flow and rule_flow_196 are the same picks read from link's flow version of default as the starting point (#80; NA on a table scored before that, or for a species whose default has no MAD floor). near_river_scope is the scope link applies the near-river pair on, read from the comparison's tuned_structure (scripts/0330): door, which puts through-wetland lines through the size test or the pair, is link's both; door_wetland, the pair on through-wetland lines alone, is link's wetland_flow. It is NA where the rule has no pair, and where compare_dir is NULL. evidence_md5 is the identity of the evidence build the rule was scored on, as the summary records it (scripts/0320; NA on a summary written before the record), which scripts/0325 holds each grade table to; curve_grad is the visit column the rule and its curve read the gradient from (segment_gradient, link's segment gradient, from #92; NA on a summary written before the record). A summary whose rule_recommended is not its tune table's recommended rule is refused: the two are written at different points of a run. A species × stage that was not fitted (no stream rule, no response, too few groups) gets one row per stage with status saying why and no values.

Usage

knw_thresholds_table(
  tune_dir,
  evidence_version,
  link_run_uid,
  link_sha,
  config = "default",
  d_floor = 200,
  compare_dir = NULL
)

Arguments

tune_dir: character(1) the directory scripts/0320 wrote (one <sp>_<stage>_summary.csv per species × stage, bound here).
evidence_version, link_run_uid, link_sha, config: character(1) the provenance every row carries.
d_floor: numeric(1) the run's shortest exception distance (m), as given to knw_tune_run(), so the 1.96 replay picks as the run did.
compare_dir: NULL or character(1) the directory scripts/0330 wrote (one <sp>_<stage>_compare.csv per compared key), read for near_river_scope; a fitted rule with a pair and no comparison there, or a comparison of another rule (tuned_rule), is an error.

Value

A list: thresholds and groups (one row per rule_id and watershed group the rule was scored on), both tibbles.

Examples

if (FALSE) { # \dontrun{
t <- knw_thresholds_table("data/tune", "knowledge/fdis/v3", "20261017_004619-3f4c713b",
                          "a17e9177", "default", compare_dir = "data/compare")
t$thresholds
} # }