ConstructEntry.nearMisses is written by every registry dimension and read by no grid

What this is

Since #285 dropped near misses from the pair alphabet, scripts/shape-registry.ts's nearMisses field (76 spellings over 35 dimensions) reaches no grid: standingGrid() reads only DELIMITER_PARTS.nearMiss through the perturbations, pairAlphabet() reads body only, and the list-run and heading-adjacency grids declare their own lists. shape-registry-pairs.ts's module comment says the opposite ("a near miss earns its place on the STANDING grid"), and #285's body says the standing grid keeps them; neither is true of OTHER_CONSTRUCTS. No census rule covers the field, so nothing fails. Measured Sep 7 by the #303 lane: line-comment's near miss ///c appears in no realized input of any grid.

The change

Either the standing grid reads nearMisses (one named row per spelling, measured for arrivals before landing) or the field is deleted with the comments that describe it, under the subtraction discipline. Measure the row count each way; prefer the deletion unless a near-miss row finds a live failure the tree's other pins do not hold (the #303 lane found two such shapes; see #331).

Related

#303, #285, #284.

Start by tracing nearMisses in scripts/shape-registry.ts and the standingGrid(), pairAlphabet(), and related grid definitions described in the issue; read the comments in shape-registry-pairs.ts and compare the findings with #331. Measure the row count for retaining versus deleting the field, then verify the chosen outcome with the census rules and the related grid cases.