Filter a long-format table to recent calendar diagonals

Returns a subset of the input data.table containing only rows whose calendar position falls within the last recent calendar diagonals of its group.

The group-wise long-form condition is rank(cohort) + dev - 1 > max(rank(cohort) + dev - 1) - recent.

Usage

.apply_recent_filter(
  dt,
  recent,
  groups = character(0),
  cohort,
  dev,
  dev_split = NULL
)

Arguments

dt

A long-format development data.table.

recent

Positive integer or NULL. When NULL or missing, dt is returned unchanged.

groups

Character vector of group columns (may be empty).

cohort

Single column name for the cohort variable (e.g. cohort).

dev

Single column name for the development variable (e.g. dev for Triangle objects, or ata_from for ATA/ED objects).

dev_split

Optional SA-boundary specifier. Accepts:

• NULL – no SA boundary; the recent wedge applies to every row.

• A single non-NA numeric scalar – the maturity target dev (= ata_to, the first CL-region dev). The recent filter is applied only to rows where dev >= dev_split (CL region); rows with dev < dev_split (ED region) are kept unconditionally.

• A data.table [groups..., dev_split] – per-group SA boundary (different k* per group). The group columns must be a subset of groups. Each row of dt looks up its dev_split via left-join; rows whose group has no matching entry (NA after the join) are treated as if dev_split = NULL for that row (recent wedge applies to all dev for them).

Value

A filtered copy of dt (class preserved), keeping only rows within the recent-diagonal window.