test(cliffs): the border residual is a WEST-edge residual (#84)

[wormeyman]

Three mechanisms have now failed on the chunk-border enrichment - the orientation-reach rival (#134), cascade double-counting (#143), and the cross-chunk destroy cascade (#149, which got 2 of 25). So this stops proposing mechanisms and describes the 23 survivors. No capture.

1. updateConnections explains NONE of them

It is the one engine pass that runs exclusively on the chunk's outer ring (applyCliffs gates it on the fifth argument of tryToAddCliff), and no audit had ever applied it - #143 excluded it deliberately as an upper bound. It therefore had the best prior of anything left for a border-shaped defect.

23 survive the cascade alone, and the same 23 survive cascade + updateConnections. A fourth mechanism ruled out.

2. The survivors are WEST

17 of the 23 sit on a chunk border, 6 are interior. Splitting those 17 by which edge, against a base rate measured over every raw border cell in the same regions rather than an assumed uniform one:

edge	survivors	expected	base rate	z
west	9	3.82	22.5%	+3.01
north	5	3.81	22.4%	+0.69
east	2	4.81	28.3%	-1.51
south	1	4.57	26.9%	-1.95

The measured base rate is what makes this a finding rather than a shape of the lattice. West carries the fewest border cells (22.5%, against east's 28.3%) and the most survivors - so this cannot be "there are simply more west cells".

Four bins were scored, so west's Bonferroni-corrected two-sided p is ~0.005. Stated explicitly, because scoring four bins and reporting the largest is exactly how a 3-sigma result later turns out to be noise.

This is a sharper localisation than the enrichment itself (z = 2.67): "on a chunk border" becomes "on the WEST edge of a chunk", which is a directional signature and therefore points at a mechanism that has a direction.

3. The lead, recorded as a lead

fixImpossibleCellsSweep clears the first clearable edge in the order L, T, R, B - west first, north second - and an edge is clearable only if it is not on the chunk's outer boundary. West + north is 14 of the 17. It is the only rule in the port whose asymmetry matches the observed one, and it is already the named cause of Nauvis's ~6% residual.

Deliberately not asserted. This spec measures where the survivors are, not why. A test needs a discriminator separating the sweep's edge order from anything else west-flavoured - and note the ore recall gap (#142) independently found all six of its cells with the nearest resource to the west, so "something is west-flavoured" already has more than one possible source in this investigation.

Verification

pnpm run verify green - 1574 app (up 4) + 13 worker + 3 container tests, 0 warnings, exit 0. Measurement only; shipping accuracy unchanged.

🤖 Generated with Claude Code

https://claude.ai/code/session_01HKjzpWJw1HbnigPP9udqN2