Borrow missing per-segment factors across segments (segment_bridged_borrowed)

Augments a per-segment factor table so every segment carries a factor for the full development range, borrowing the entries a segment cannot estimate from a donor segment that can. Used only by the "segment_bridged_borrowed" treatment: early-development factors stay regime-specific (each segment's own estimate), while late-development factors a segment never reaches are filled from another segment.

Donor rule (“recent”): for each (group, dev) the donor is the segment with the largest segment_id that has a non-NA primary factor at that dev – i.e. the most recent regime whose own cohorts developed that far. The bridged band guarantees a donor exists at every dev that any segment needs (the boundary factor gaps are closed by the bridge), so no dev is left unfilled.

The borrow only adds rows for (segment, dev) combinations the segment lacks; rows a segment owns are never overwritten.

Usage

.borrow_segment_factors(sel, groups, dev_col, factor_cols)

Arguments

sel

A data.table of per-segment factors with columns groups, dev_col, segment_id, and factor_cols (plus optional ata_to / ata_link carried along).

groups

Character vector of group columns (may be empty).

dev_col

Single column name holding the development index ("dev" after the projection-time rename).

factor_cols

Character vector of factor columns to borrow. The first element is the primary factor ("f_sel" / "g_sel") whose non-NA presence defines whether a segment owns a dev.

Value

sel augmented with borrowed rows so every (group, segment, dev) present in the donor space is covered.