Assume matching corrections, R=28, m=2, Z=0 and exactly one doubled helper, with nonzero coordinate flow.

## 1. Complete nonzero-flow budget cover

Suppose flow is nonzero. If the D is outside group M, its array is
nonzero in addition to the three untouched M-arrays; A>=4 then gives
R>=19+4+6=29. Therefore D is in the block (M,U), with U an M-edge.

If a!=1, none of the three M-arrays cancels, so A>=3 and Pi<=19.
Equality forces Pi19,A3,C6,X0. The private line budget is
sum f=2(Pi-1)=36. Four inactive centers cost at least five and the
two active centers at least eight. Therefore their costs are exactly
f8,f8, with four inactive f5, and the proved f8 law forces a=1/2.

If a=1, both active Gamma values equal1/2, so f>=9 at each and
Pi>=20. Since A>=2, equality forces Pi20,A2,C6,X0, with both active
costs f9 and all four inactive costs f5.

These are the only two nonzero-flow corners. No coefficient grid or
sign assumption is used to reduce the originally arbitrary nonzero a.

## 2. The a=1/2 corner: three flow blocks cannot meet the demands

The two active centers are glue-free, so their B rows are zero. Every
inactive f5 center has one nonzero-marginal line of marginal -1/2;
its B row has two entries one and otherwise zero. Hence H0>=0,
H0_U=0, every edge touching U has H0<=1/2, and every other edge has
H0<=1. These conclusions follow from this equality budget, not from a
general positivity premise on private coefficients.

Write M={U,V,W}. The sole beta value is one in block (M,U). In F the
pair-node demand is

    demand_e=H0_e+1[e=U]+(1[e=V]+1[e=W])/2-2*1[e in M].

All three M nodes have strictly negative demand: U has demand -1,
and V,W have demand at most -1/2. Off M the demand is H0>=0.

C6 allows exactly three nonzero-flow blocks, and their divergence
has at most six nonzero pair-nodes. At Z=0, projection says
sum_B delta_B*(1_W-1_V)=0. Summing F over edges incident to a coordinate
x therefore gives

    sum_(e containing x) H0_e = 1       if x belongs to U,
                               3/2     otherwise.

The offmatching incident H0 sum is consequently one at each U endpoint,
and at least1/2 at each other coordinate. The positive offmatching
edges cover all six vertices and hence number at least three. Together
with the three negative M demands this saturates the maximum of six
demanded pair-nodes. There are exactly three positive offmatching edges;
they form a perfect matching M' disjoint from M.

At an endpoint u of U, its unique M' edge would need H0 value one to
provide the entire offmatching sum. Its other endpoint is inactive,
because the only other active center is u's M mate. But every such edge
has H0<=1/2. Contradiction.

## 3. The a=1 corner: the private split marginals disagree

Both active centers u,v have a single nonzero helper class of value one
and cost f9. There are exactly two main N=-1 on together lines for U,
one F4 and one F5. All glue is on the F5 line with marginal1/2.
Its glue vector must be (-1/2,1/2,1/2) up to position permutation.
Indeed a one-glue F5 line with N=-1 would have marginal one, whereas
the no-star three-glue form has marginal1/2. Thus there are two positive
glues and one negative glue. This is also the proved f9 equality law.

Fix u's F5 line. Its half is {u,v,w}, where w is inactive, since every
main line is together for U. All three opposite centers are inactive.
The three nonzero u-glues activate their unique f5 lines in this split,
each of marginal -1/2.

Choose a positive u-glue. At its opposite center the N-free f5 line
has marginal -1/2 and contains a glue +1/2. It must therefore be the
three-glue form (1/2,-1/2,-1/2), not a single -1/2 glue. Its other two
nonzero glues activate the v and w lines in the same split. The v line
is its sole glue-bearing line and so has marginal +1/2. The w line is
its unique f5 line and has marginal -1/2.

The sum of glue marginals on the {u,v,w} half is consequently

    1/2+1/2-1/2=1/2,

whereas the opposite half has total -3/2. Both totals sum the same
three-by-three glue matrix and must agree. Contradiction. This corner
is excluded by the private rows alone, without a zero-flow assumption.
