Resolve a benefit threshold from development action scores

Resolves a fixed signed threshold or the weighted development-sample average treatment effect reference. The result is a hypothetical outcome-scale benefit threshold, not a measured economic cost.

Usage

margot_policy_value_threshold(
  dr_scores,
  weights = NULL,
  value_threshold = 0,
  threshold_multiplier = 1,
  treatment_column = NULL,
  control_column = NULL
)

Arguments

dr_scores

Numeric matrix of finite, original (unweighted) binary action scores, on a common outcome scale where larger scores are preferred. Recognisable control and treatment column names are required unless both column indices are supplied.

weights

Optional finite non-negative development weights, aligned with rows and with positive total weight. Weights define the averaging population and are applied once.

value_threshold

A finite numeric scalar (default zero, preserving the original objective), or "ate" to use the weighted mean treatment-minus-control action-score contrast in these development rows.

threshold_multiplier

Finite non-negative scalar multiplying the fixed threshold or development ATE; default one. Multipliers are relative references, not guaranteed extreme ranges.

treatment_column, control_column

Optional distinct integer column indices identifying the two actions. Supply both together.

Value

A margot_policy_value_threshold list with resolved value, source, multiplier, development_ate, development row and weight counts, and action-column indices. Preserve this object unchanged when evaluating the associated learned rule on separate observations.

Details

Subtract the resolved value from the original treated action score before applying the analysis weight. Apply this same objective to every comparator. Original treatment effects retain their signs: an effect below an ATE reference need not be harmful. A zero or negative ATE is a relative reference and cannot automatically represent a positive treatment expense. This helper does not establish causal identification or independence of the supplied scores; callers must keep nuisance fitting, threshold estimation and rule learning within their declared development boundary.