Assume matching corrections, R=28 and Z=0. The doubled-helper sources are a multigraph on the active centers.

## 1. A tight separator with two classes on the same source

**The same-prefix cost-thirteen lemma.** Let u have two distinct nonzero helper classes with
the same source pair U containing u, total line cost f(u)=13, and a
nonzero marginal on a line separating U. Then there is exactly one
nonzero-marginal separating line, its marginal is -1/2, all together
lines have zero glue and zero actual stars, and the two class values
sum to zero. Additional balanced classes are allowed in the hypothesis
but cannot occur at this cost.

The two nonzero classes require at least four N, all on together lines
for U. A nonzero-marginal separating line costs at least five, including
the proved collision-safe scope. Six N together with a required glue
and an actual star already cost at least 12+1+2=15. Five N can only
come from these two classes, since another balanced class requires two
additional N; they would cost at least ten on together lines, plus five
on the separator. Thus there are exactly four N, all on together lines.

Their N cost eight and the separator costs at least five. Equality f13
forces the together lines to have neither glues nor actual stars, and
leaves precisely one N-free F5 separator. Every actual star at u is on
that separator, with total mass one. The N-free line formula gives
star sum -2r, so its marginal is r=-1/2. The total Gamma is also -1/2,
hence the sum of helper class values is zero. This argument uses actual
star mass, without assuming star positivity.

Consequently, at a same-source two-class center with f13, its total
separating marginal is either zero (if no such line exists) or -1/2.

## 2. Two attachment vertices: d>=2 is impossible

All D source pairs equal K. With d=2, both endpoints have two nonzero
classes, so the proved f13 sharing lemma and four inactive costs give

    Pi >= (13+13+4*5)/2+2 = 25.

If K is not an M-edge, both D blocks are outside M and A>=5, already
too large. If K is an M-edge, at most one D is in its M-block; the
other is outside. Two other M-arrays remain unchanged and the outside
array is nonzero, giving A>=3. R28 therefore forces Pi25,A3,C=X0.
Both active endpoints have f13 and all inactive vertices have f5.

The flow is zero, and all nonzero helper classes have source K, so
source balance requires NT_K=w_K-1=1. The same-prefix cost-thirteen lemma restricts each
endpoint's separating contribution to {0,-1/2}. Their sum cannot be
one. This excludes the exact tight corner without applying the
single-source 403-column dual beyond its allowed D-column scope.

For d>=3, the 2d N at each endpoint and one actual star give the proved
charge Pi>=12+5d. At most one of these distinct same-source D blocks
is in M. The two other M-arrays and at least d-1 outside arrays give
A>=d+1. Thus R>=13+6d>=31. These estimates allow arbitrary extra loops
and mixed columns; adding them only consumes support.

Thus the Z=0, nonzero-helper m2 frontier has exactly one D.

## 3. Three attachment vertices

At Z=0, assign pi(x)=f(x)/2+r_x/2, where r_x is D-degree. Inactive
centers cost at least 5/2, while active degree costs are at least

    J(0)=J(1)=9/2,  J(2)=15/2,  J(r)=1+5r/2 for r>=3.

These bounds permit balanced N and are nondecreasing in r. At most one
M-edge lies wholly in K, so at least two M-arrays remain untouched:
A>=2.

Two D with the same source give degrees (2,2,0). The two f13 endpoints,
one helper-free active f9 and three inactive f5 yield Pi>=27. At most
one of the D blocks is in M, so A>=3 and R>=30.

For three D on three vertices, allowing parallel sources in distinct
blocks, the only unordered degree patterns and their charges are

| Degrees | Pi lower bound including three inactive centers |
|---|---:|
| (2,2,2) | 30 |
| (3,2,1) | 28 |
| (3,3,0) | 29 |

For more D, select any three columns. The actual degrees are at least
the selected degrees and J is nondecreasing, so Pi>=28 persists. With
A>=2 this excludes every d>=3 case at R28.

The remaining nonzero D support at m3 is one D, or two D with different
sources forming a path. No closure of those cases is asserted here.

## 4. Four attachment vertices

The active/inactive baseline is Pi>=23. Each center with D-degree at
least two adds at least three by the same charge. Two such centers give
Pi>=29 and are already excluded at R28.

With two D sharing exactly one endpoint, the degree charge gives
Pi>=26. At most one source is an M-edge, and at least one helper block
is outside M. At least two M-arrays remain unchanged and the outside
array is nonzero, so A>=3 and R>=29. Two D with identical source have
two degree-two centers, already excluded.

For d>=3, the only graph on four vertices with at most one vertex of
degree at least two and at least three edges is the three-edge star:
its degrees are (3,1,1,1). Indeed all other vertices have degree at most
one, so there can be at most three edges incident to the exceptional
vertex, and no fourth edge can be added. This includes multigraphs:
parallel edges would make both their endpoints have degree at least two.

The star has Pi>=5+J(3)+3J(1)=27. At most one of its three sources
can be an M-edge. Therefore at least two D blocks are outside M and
at least two M-arrays remain nonzero, giving A>=4 and R>=31. All other
d>=3 configurations have at least two high-degree centers and Pi>=29.

Thus the Z=0, nonzero-helper m4 frontier has one D, or two D with
disjoint sources. The proof did not assume their coefficients' signs.
