Skip to contents

Per-cell effective dev_min for the segment mini-triangle band

Source: R/utils.R
dot-compute_segment_mini_tri_bounds.Rd

For each cell, returns the minimum dev that keeps it inside its segment's fit mask. The mask is the union of two regions:

  1. The segment's natural mini-triangle: dev >= max_cal - seg_last + 1.

  2. A bridge extension along the calendar diagonal anchored at the next (newer) segment's first-cohort midpoint dev. The bridge lets each older segment connect to its successor, filling the late-dev cells of its early cohorts that would otherwise be cut by the natural mini-triangle wall and leave projection cells unreachable.

Bridge construction (segments ordered by segment_id, lower id = older cohorts). For each segment s except the newest (no successor), find segment s+1's

  • first_rank – cohort rank of s+1's first cohort,

  • seg_dev_min – max_cal - last_rank(s+1) + 1,

  • first_cohort_dev_max – max_cal - first_rank(s+1) + 1,

  • mid_dev – floor((seg_dev_min + first_cohort_dev_max) / 2).

The bridge diagonal for segment s is at ext_cal_idx(s) = first_rank(s+1) + mid_dev(s+1) - 2 (the cell one cohort earlier than s+1's first cohort, at the same dev as that first cohort's mini-triangle midpoint). Each cell in segment s then takes

effective_dev_min = min(seg_dev_min(s), ext_cal_idx(s) - coh_rank + 1)

with pmin(..., na.rm = TRUE) so the natural mini-tri wall stays put when no bridge applies (the last segment, or cells whose bridge ray lies above the wall).

Bridges do not cascade: segment s is bridged only from segment s+1, not from s+2. The bridge only ever widens a segment's mini-triangle.

Usage

.compute_segment_mini_tri_bounds(coh_ranks, seg_ids, max_cal, bridge = FALSE)

Arguments

coh_ranks

Integer vector. Per-cell cohort rank within the group.

seg_ids

Integer vector. Per-cell segment id (1 = oldest).

max_cal

Integer scalar. Maximum calendar index in the group.

bridge

Logical. When TRUE (used by both segment treatments), widen each older segment's mini-triangle with the calendar-diagonal bridge anchored at the next segment's first-cohort midpoint dev. When FALSE, return the natural mini-triangle wall only (no boundary-gap closure); retained for diagnostics and tests.

Value

Integer vector (same length as coh_ranks) of per-cell effective dev_min for the mini-triangle filter (bridged or pure).