Skip to contents

Construct an sda_anchor object

Source: R/sda_anchor.R
sda_anchor.Rd

Low-level constructor. Prefer as_sda_anchor when converting from an sda_fit. Use this constructor when building an explicit / manual anchor from pre-specified fields (e.g. from a published attribute ordering).

Usage

sda_anchor(
  anchor_type = "explicit",
  source_class = NULL,
  source_call = NULL,
  group_levels = NULL,
  selected_attributes,
  candidate_universe = NULL,
  stage_table,
  branch_candidate_map = NULL,
  removal_history = NULL,
  weights_used = FALSE,
  weight_summary = NULL,
  loo_mode = NULL,
  mc_iter = NULL,
  mc_seed = NULL,
  mindenom = NULL,
  alpha = NULL,
  stop_reason = NA_character_,
  reproducibility_notes = character(0),
  canon_notes = character(0),
  task_hook = .sda_anchor_task_hook()
)

Arguments

anchor_type

Character scalar: "sda_fit" for anchors derived from a fitted sda_fit object, or "explicit" for manually-declared anchors.

source_class

Character vector: class of the source object, or NULL.

source_call

Language object or NULL: the call used to produce the source object.

group_levels

Integer vector of class/group levels, or NULL.

selected_attributes

Non-empty character vector of selected attribute names in stage order.

candidate_universe

Character vector of all attributes evaluated, or NULL.

stage_table

Data frame with at least columns stage_id and attribute.

branch_candidate_map

Named list for SORT branch-level candidates, or NULL (reserved for future SORT).

removal_history

List of per-step removal records, or NULL.

weights_used

Logical. FALSE unless weighted SDA is available.

weight_summary

List or NULL.

loo_mode

Character scalar or NULL: LOO mode string from SDA settings.

mc_iter

Integer or NULL: MC iterations from SDA settings.

mc_seed

Integer or NULL: RNG seed from SDA settings.

mindenom

Integer or NULL: MINDENOM from SDA settings.

alpha

Numeric or NULL: significance threshold from SDA settings.

stop_reason

Character scalar or NA: SDA stop reason.

reproducibility_notes

Character vector.

canon_notes

Character vector.

task_hook

List. Machine-readable metadata for future agent/pipeline consumers. Defaults to the standard anchor task hook (see ?sda_anchor).

Value

Object of class c("sda_anchor", "list").

Details

An sda_anchor is a typed structural object that carries SDA selection history for future SORT (staged CTA) workflows. It is not a fitting object and does not estimate propensity scores.

What an SDA anchor is not:

  • It is not a propensity-score estimator. SDA produces stage order and selected attributes, not a propensity stratification.

  • It is not an implementation of SORT or GORT. Both remain future reserved workflows.

  • Explicit / manual anchors are not SDA-derived and must be labeled anchor_type = "explicit".

Task hook: The default task_hook marks implementation_status = "anchor_only_no_sort", lists prohibited_downstream = c("propensity_weighting", "fraud_demo"), and requires human review.

See also

as_sda_anchor, validate_sda_anchor, sda_fit