The initial R29 map reduces this branch to two disjoint D sources U,V,
with coefficients a,b. One D is excluded by E23, or U25 when its M
array cancels; repeated/intersecting sources and d>=3 are excluded by
their actual R29 degree/array costs. Pi>=23, A>=1, and nonzero flow
would cost C>=6, exceeding29. Thus coordinate flow is zero.

Write q,r for the two inactive vertices. The exact source equations are

    w_U=1+NT_U+b*[U is a side of the V block],
    w_V=1+NT_V+a*[V is a side of the U block].             (T2)

Here NT is the sum of separating marginals at the two source endpoints.
All mixed terms vanish in these equations. A nonzero separator costs
at least13. A unit-class cost13 separator has marginal1/2, with all
other marginal lines at that endpoint zero: its two main lines cost4
each and its separator costs5. This statement allows balanced collisions.

## 1. Local incidence and glue bounds

**J12.** At a sole unit-class center of cost f<=12, all glues and actual
stars lie on its two main together lines, even with balanced N.

The two main N occupy distinct together lines. Five N plus a required
star leave no glue at cost12, contradicting Gamma=1/2. Three main N
occupy three lines. Four main N occupy four lines; a minimal main pair
plus a balanced pair can occupy two or more. Each N-line costs at least4,
and an F4 line has zero marginal. Three N-lines at total cost12 would
all be F4 and give Gamma0, so only two N-lines fit. An additional N-free
line costs at least5, exceeding12. All stars therefore remain on the
together lines. Their source pairs contain the class mate and cannot
be either U or V. The same two-line confinement holds for a half-class
center of cost at most10, simply because three N-lines already cost12.

**G11.** All glues at a unit-class center of cost at most11 have
magnitude at most1, and belong to{0,+/-1/2,+/-1}.

At f9 they have magnitude1/2; f10 is impossible. At f11, four N force
one glue, equal to Gamma1/2. Three N cost at least12. For two main
N=-1 on the two lines, the exact six-variable affine check covers all
456 faithful-support masks of cost11. Its eight surviving forms have
precisely the stated glue-value set. Their equality equations and
coefficients are saved; this is an unrestricted rational local check,
not coefficient sampling.

Every star in block(M,U) is centered outside U, and each outside center
has exactly one star column in that block. At an inactive center, such
a star lies on half U plus that center. These literal facts are checked
for all45 blocks.

Two inactive F5 centers together contribute at most1 to a fixed block
(M,U) whose source is disjoint from them. Each individual contribution
is0,1/2 or1. A contribution1 means a dense F5 line on half U+q. Its
glue to r forces r's unique nonempty line onto the complementary half,
whose source is V, so r contributes zero to(M,U). Otherwise both
contributions are at most1/2. This proves the bound without constraining
the other centers' coefficients.

## 2. Pi<=25: all sixteen forced equality forms are excluded

The doubled line budget is at most46. No endpoint can reach separating
cost13 while the other three cost at least8 and the inactives cost10.
Thus NT_U=NT_V=0. T2 forces both side incidences, hence both D belong
to the unique matching containing U,V. It also forces a,b in{-1,1}.
A negative class costs at least10, so Pi<=25 forces a=b=1, their group
M, Pi25, four active f9 centers and two inactive f5 centers.

The four active glue lines are dense and in a common split with U and
V on opposite sides. A third vertex from the other D pair would force
a separating glue at its opposite f9 mate, which is impossible. All
four dense lines are then forced by their crossing glues. Relabel the
split as014|235, U=01,V=23. The literal main-N positions force

    L=(1/2)*[[1,1,-1],[1,1,-1],[-1,-1,1]].

At each active center, the dense-line N is fixed. Its other h side
has exactly two possible N, one of which must be chosen. No other
private support fits. These give exactly2^4=16 equality forms. The
other split and all labels are equivalent under permutations of the
inactive pair and the matched pairs. This covers all relative N choices.

The checker derives these16 cores from literal incidences. Twelve have
five nonzero B demands, two have six, and two have four. All require
A>=4; R29 thus forces A4,C=X0. For each four-demand core every A must
lie in one of those four blocks, giving16 allowed columns. All16 are
allowed simultaneously in the exact affine equations. Rational row
duals prove rank16 and force a unique solution with six nonzero A.
Each dual is stored with its original literal/array row labels and
exact weights and is checked by multiplication. Thus none of the16
forms admits A4. No fixed-core conclusion is extrapolated to other
cores; all equality forms have been covered explicitly.

## 3. Pi26

A<=3 forces at least one D block into M. If one is outside, the inside
D must be a unit, since otherwise four arrays remain nonzero.

If both coefficients are nonhalf, all endpoints cost at least9.
A separator would require line cost at least13+27+10=50, beyond48.
Thus both NT vanish. T2 forces a common matching group, and the inside
block makes that group M; consequently a=b=1.

If both coefficients are halves, both blocks must be in M since no
M-array cancels. T2 requires NT_U=NT_V=1/2, hence one separator at
each pair. Even the weaker cost13+8 at each pair gives total52>48.

Suppose only b is a half. Its source cannot have a separator: the
required total would be at least13+8+18+10=49. If both blocks are in
M, T2 forces a=1. If the half block is outside M, the other is already
a unit in M. A half source outside the M matching would give w_V=1
from T2, impossible; if V is an M-edge, its outside group cannot also
contain U. Then NT_U=1 is required, whereas a unit endpoint at this
budget has cost at most13 and NT_U is0 or1/2. This excludes the outside
half block.

In the remaining in-M unit/half case, NT_U=1/2 forces the exact costs
13,9 on U,8,8 on V and5,5 inactive. The cost13 separator has marginal
1/2 and is F5. A dense separator would cross the f9 U-mate on a
separating line, impossible. Its single positive glue must therefore
meet an inactive F5 line. That line must be dense. It forces every
center opposite it to carry glue, including a glue-free V endpoint:
the opposite half contains the separating U endpoint but excludes its
mate, and cannot exclude both V vertices as well. Contradiction.

It remains a=b=1 in M. At Pi26 every endpoint costs at most11, so J12
excludes its stars from both D blocks. An inactive f7 would have the
other five centers at their minimum costs; all its crossing glues would
have magnitude1/2, making its supposed F7 line F5. Hence both inactive
centers are f5, and each D block has actual star sum at most1. Both
therefore require strictly positive alternating sums. With A<=3 the
nonzero remaining M-array consumes one direction; outside-M directions
can only be an opposite-coefficient pair in one zero-total array.
That pair cannot supply positive sums to two different blocks, and
group-M alternatings belong to neither D block. This excludes Pi26,
including its possible idle-loop or mixed slack.

## 4. Pi27 with a nonunit second coefficient

A<=2 forces both D in M and at least one unit, say a=1. If b is
nonhalf, each U endpoint has cost at most13. There is at most one
nonzero separating line at U, so NT_U is0 or1/2. T2 then forces b=1
or1/2. Thus a genuinely nonunit b must be1/2.

For b=1/2, NT_U=1/2 requires a unit endpoint of cost at least13.
The V-endpoint cost sum is at most18, hence one has f8 and is
glue-free, and the other has cost at most10. Their stars stay on
together lines for V and contribute nothing to block(M,U). Neither
U endpoint can have source U. All A are in M, so the U block requires
actual star sum2 solely from the inactive centers.

Each inactive costs5 or7. A star with source U needs half U plus that
inactive vertex. Its opposite half contains the glue-free V endpoint.
Thus neither a dense F5 nor an F7 line can occupy that half. Only a
sparse F5 remains, contributing at most1/2 to the particular block.
The two inactives supply at most1, contradicting the required2.

The remaining Pi27 case and every Pi28 case have two unit D in M.

## 5. Two unit D in M, Pi<=28, A<=2

All A are in group M: its remaining nonzero array needs one direction,
while an outside zero-total array would need two more. Thus both D
blocks require actual star sum2, for combined sum4. Loops and mixed
columns vanish in these B aggregates.

The minimum doubled private line cost is46, with four unit f9 centers
and two inactive f5 centers. At Pi<=28 the excess is an even integer
at most6. Unit f10 and inactive f6 do not occur.

If a unit center has cost at least13, it is the only such center.
It has cost13 or15; cost14 would leave an impossible one-unit excess.
All other unit centers have cost at most11 and contribute no stars
to either D block. An inactive f7 could then have only one partner
with non-half-magnitude glues. But its N-free F7 form(r,t,-t), r=-1/2,
needs two such partners because t is neither0 nor+/-1/2. Therefore
both inactives are f5. In the D block whose source contains the high-cost
unit center, neither endpoint contributes, the other unit centers
contribute zero, and the two inactives together contribute at most1
by section1. The required sum2 is impossible.

Otherwise all unit centers cost at most12 and J12 excludes all their
stars from both D blocks. We bound the combined contribution of the
two inactive centers, preserving possible negative stars and nonempty
zero-marginal lines. The even cost excess gives the following exhaustive
classification, also checked as99 small integer cost profiles.

- If an inactive has f11, every other center is at minimum cost. All
  its glues therefore have magnitude1/2. Its only possible line costs
  are one nonzero-marginal F5 plus one zero-marginal F6: a single
  N-free line costs at most9; two nonzero lines cost10 and cannot gain
  just one unit. Its total positive star mass is at most1+1/2.
  The other inactive contributes at most1, giving at most5/2<4.
- An inactive f9 has one F9 line. If the other inactive is f7, all
  unit centers are f9, making that F7 impossible by the two-partner
  argument. Otherwise the other inactive is f5 and at most one unit
  center has f11. Two of the F9 line's three nonzero glues have
  magnitude1/2, and the remaining glue lies in{+/-1/2,+/-1} by G11.
  Sum-1/2 forces that remaining glue also to have magnitude1/2,
  making the line F5, not F9.
- An inactive f8 is accompanied by exactly one other excess3, either
  a unit f12 or another inactive f8. Its F8 line has two nonzero
  glues, at most one with an exceptional partner. The normal glue is
  +/-1/2; sum-1/2 forces the pair(+1/2,-1). Its stars are(1,-1/2,1/2),
  up to permutation, so its contribution to the particular source
  block is at most1. Each inactive contributes at most1, totaling2.
- Otherwise both inactives have cost5 or7. When an f7 occurs all
  unit centers have cost at most11, since an excess3 together with
  excess2 cannot complete the even budget at most6. Among the two
  glues t,-t of an F7 line, at most one meets the other inactive;
  the other meets a unit center. G11 gives |t|<=1. Its largest actual
  star is at most3/2. Its single line has one source pair and at most
  one star in the corresponding M-block. Thus the two inactive
  contributions total at most3, below4.

Inactive f10 cannot occur: its excess5 would need an unavailable
excess1. These cases exhaust the budget and contradict the combined
star demand. This completes every two-unit case and hence every
Z-free, helper-present m4 case at R<=29.
