Frozen before testing the height2 atom filters, 8 November 2026.
Motivation: Wren independently verified the height1 inverse and reports that
height2 heavy Fourier tails survive random multipliers. We need actual
arithmetic control, not a refined typical-frequency heuristic.

Derived hypothesis to test: the six entry families can be partitioned into
2 incoming classes (previous defect zero versus positive) times3 outgoing
patterns (d_i+2,d_i+3)=(0,0),(1,0),(1,1). The mandatory d_i=d_i+1=0
comes from the genuine phase boundary. At a completed height2 clip, use
gamma=2^(B_i-2)3^-(i+4). Every profile source shift is an integer multiple
of gamma, independent of the remainder of the word.

Generate exact Laurent filters for these six disjoint atoms. Look for
unit-circle zeros, especially the high-incoming/(0,0)-outgoing atom.
This is a test of the derived filter algebra, not a registered statistical
claim and not a prediction of all-depth moments.

Check finite actual source indicators on lambda11/7,height2,b8, rowi2,
phasec=1+{-i lambda}, common binary period<=2^17. Repeat on fixed terminal
total slices. Directly measure whether any actual atom excites a frequency
at which its own filter vanishes. A zero on arbitrary functions is not by
itself evidence of a hidden actual trajectory component.

CPU bounded120seconds, no GPU, paid compute or subagents. No new broad
random sweep or paper submission until this mechanism has been extracted.
