Assume matching corrections, R=28, Z=0 and m in {3,4}, after the source-degree exclusions.

## 1. Three local tools

### E23: one-source endpoint cost

Suppose exactly one D=a has source U={u,v}, flow is zero, and any
additional N have zero helper classes. The proved source balance says
NT_U=w_U-1, equal to1 or-1, allowing balanced N at any center. Then

    f(u)+f(v)>=23.                                           (E23)

For a!=1/2, both Gamma=a-1/2 are nonzero. A nonzero separating marginal
costs at least13 at one endpoint and the other costs at least9. If the
sum were at most22, the costs would be13,9. The f9 law forces a=1,
whereas the f13 separator contributes Gamma=1/2, not NT=+/-1.

For a=1/2, an f13 separator would contribute Gamma0. Thus a nonzero
separator costs at least14, and the other endpoint at least8. At total
22 the costs must be14,8. Three main lines or an extra together line
would already cost12 before a separator5. There are exactly two main
N=-1/2 together lines, with total cost at most9. Their F4/F5 marginal
sum is0,1/4 or1/2, never the required -NT=+/-1. This proves E23.

If a=1 and U is an M-edge, the previously recorded U25 bound improves
this to f(u)+f(v)>=25. Its proof is likewise local and permits balanced
N elsewhere; it is not confined to m2.

### B10: a helper-free active center of cost at most10

If every zeta at an active center is zero and f<=10, all actual stars
are nonnegative, every positive glue is at most1/2, and f is9 or10.
At f9 the earlier B9 classification gives B entries zero/one and all
nonzero glues have magnitude1/2.

To prove the new f10 assertion, Gamma=-1/2 requires a glue and actual
star mass1 requires a star. Balanced parity leaves exactly two N: four
N plus a glue and a star cost at least11. They form one balanced pair,
with coefficients n,-n on distinct lines. Each line costs at least4;
no third nonempty line can fit. At total10, F4+F6 has marginal0, so
the only possibility is F5+F5.

A one-N F5 is either a one-glue line of marginal -n with one actual
star n, or a three-glue line of marginal -n/2 with no actual stars.
Two of the first kind have star sum0, and two of the second kind have
no stars. Thus there is one of each. The star-bearing line must have
n=1, so its sole glue is-1. The other line has N=-1, marginal1/2,
glues(-1/2,1/2,1/2), and no stars. The sole actual star is+1.
The exact affine-mask check permits arbitrary rational n and glues.

### P14: positive two-class centers of cost at most14

Suppose a center has two nonzero classes with distinct source prefixes,
values (1/2,1/2), (1,1/2), or (1,1), and f<=14. Then every private marginal
at that center is nonnegative, and every B entry is nonpositive.

Each class requires two N. Six N plus a required actual star cost14
and leave no glue; this would give Gamma0, and the glue-free actual
star mass would be twice the sum of the two class values. In the
applications below those values are (1/2,1/2), (1,1/2), or (1,1), so this
cannot equal1. Equivalently their Gamma is strictly positive and a
glue is required, making six N cost at least15.

Five main N would use at least four lines: distinct-prefix class line
sets share only one line. Their cost is at least16. Extra balanced
classes require two N. Therefore in these applications there are
exactly four main N, on exactly three lines, with all specified N
coefficients negative. Every line costs at least4 and hence at most6.
F4 and F6 have marginal0; the collision-safe F5 law with any specified
negative N gives positive marginal. There is no room for another line.
Thus every marginal is nonnegative. The assertion P14 is used here
only for the three positive value pairs explicitly listed.

## 2. One D at m4

The other two active centers have zero helper classes and cost at least9
each; the two inactive centers cost at least5 each. E23 gives

    Pi>=ceil((23+18+10)/2)+1=27.

One D leaves at least two M-arrays, so A>=2. This gives R>=29 and
excludes m4,d1, including mixed or idle-loop slack.

## 3. One D at m3

The third active center w is helper-free with f(w)>=9, and inactive
cost is at least15. E23 gives Pi>=25. A D outside M needs A>=4 and
is impossible at R28. Hence D lies in (M,U), U={u,v} in M.

### 3.1 Nonunit a

A>=3 leaves only Pi25,A3,C=X0. The endpoint cost is at least23, so
every inactive f<=48-23-9-10=6 and hence is5. There is an M-edge W
with two inactive endpoints. Its F equation gives H0_W=2-a, while
the f5 endpoints give H0_W in{0,1/2,1}. Thus a is1,3/2 or2. The
nonunit values make each endpoint cost at least11. The total48 then
forces endpoint costs13,11, third active f9 and all inactive f5.

The f13 separator is the only nonzero separator; NT=1 forces its
Gamma=a-1/2 to equal1 and hence a=3/2. Its F5 glues have magnitude1.
The other endpoint cannot have any separating glue at f11. All other
centers, including the helper-free f9 center by B9, have glue magnitude
1/2. The separator has nowhere to cross. Contradiction.

### 3.2 Unit a=1

U25 gives Pi>=ceil((25+9+15)/2)+1=26. Since A>=2, the only budget is
Pi26,A2,C=X0. All alternatings belong to M: an outside zero array needs
two columns in addition to the two nonzero M-arrays. Thus the D block
has no block-alternating column and needs actual star sum2.

The total line cost is50. Every inactive f is5, because U25 leaves at
most one excess unit. The only endpoint/third-center cost pairs are

    f(u)+f(v)=25, f(w)=10; or 26, f(w)=9.

In the first pair, both endpoints cannot be separating, since that
costs26. A unique f13 separator supplies1/2 rather than NT1; a unique
f14/f15 separator is excluded by the same nonnegative main-marginal
argument used for U25. Unit-class f10 is impossible. Thus the only
endpoint pair left is16,9, with a single separator of marginal1 at
the first endpoint. Its glue to the f9 mate is zero, since the mate's
glue lines are together lines. Each of the other two entries is at
most+1/2: inactive f5 has that bound, and the third active center does
by B10. To sum to1, the separator must have exactly two glues+1/2.
Its effective star vector has three nonzero positions. Each N can
cancel at most one, so glue cost plus twice(actual stars+N) is at
least2+2*3=8. The two main lines then have total cost at most8 and
zero marginal, contrary to their required marginal -1/2.

In the second pair, the helper-free center has f9, so B9 gives its
B entries zero/one and nonnegative stars of total mass1. All inactive
B rows also have entries zero/one. The two M-edges other than U have
H0=1, forcing both directed entries1. Every inactive center's half
therefore contains its M-mate, and none of its stars can have source U.
Neither u nor v can have source U. Only w can contribute to the D
block, and its total nonnegative mass is1, less than the required2.
This closes every m3 single-D case.

## 4. Two D sources forming a path at m3

The shared endpoint has f>=13 by the proved two-class lemma, each
leaf f>=8, and inactive cost15, so Pi>=24. Two outside-M D need A>=5
and cannot fit. One D is therefore in (M,U), coefficient a, and the
other, coefficient b, has a different intersecting source and lies
outside M. Write U={u,v}, the other source{u,w}, with u shared.

If a!=1, A>=4 forces Pi24,A4,C=X0. Equality gives shared f13, both
leaves f8 and every inactive f5. The leaf laws force a=b=1/2. P14
makes B_uv<=0; the glue-free leaf v has B_vu=0. But F on U, which
cannot be a side of a block with intersecting source{u,w}, needs
H0_U=2-2a=1. Contradiction.

If a=1, the v leaf costs at least9 and Pi>=25. The exact budget is
Pi25,A3,C=X0. Its line sum46 leaves only two active cost patterns:

    (f(u),f(v),f(w); b)=(14,9,8;1/2), or (13,9,9;1).

Indeed inactive f6 and unit-class f10 are impossible, and an f8 or
nonzero-Gamma f9 leaf forces the displayed value. P14 applies in both
patterns and gives B_uv<=0. The unit f9 leaf has B_vu=-1 because
all of its marginal1/2 is on a together line. Hence H0_U<0, whereas
F requires H0_U=2-2=0. This excludes every path case.

## 5. Two disjoint D sources at m4

Let sources be U,V and coefficients a,b. Each endpoint has just one
nonzero class. There are two inactive centers. Thus Pi>=23. At Pi<=25,
the line sum is at most46, only four above the baseline active32 plus
inactive10. No endpoint can reach separating cost13, so all separating
marginal sums NT_U,NT_V vanish.

The source equations obtained directly from F are

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

Their own helper contributions cancel 1-2a or1-2b from the endpoint
marginals. Since each w is0 or2, both side incidences must hold and
a,b belong to{-1,1}. Two disjoint sources determine a unique matching,
so both D are in the same group. Value1 needs f>=9 at each endpoint;
value-1 needs f>=10 by the proved f<=9 laws. Pi<=25 therefore forces
a=b=1, their group M, Pi25, all four active f9 and both inactive f5.

### 5.1 The forced Pi25 private core needs A>=4

Every active f9 glue line is dense, with marginal1/2 and glues
(-1/2,1/2,1/2), and its half contains its own D source. Its third
vertex cannot be an endpoint of the other D: the opposite active
mate would then receive a glue on a separating line, impossible at f9.
Thus the third vertex is inactive. The opposite half consists of the
other D source plus the other inactive vertex. The dense opposite
active rows then activate every row in this partition.

The main N on a together half has its V position opposite the M-mate
of that half's third vertex, a literal incidence checked for all360 N.
The two active rows and two active columns consequently have negative
glue at the inactive coordinate. Ordering the inactive coordinate last,
the forced matrix is

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

The two inactive stars are+1, one in each D block. All active F5 stars
are zero; their remaining F4 stars have sources containing their own
mate and contribute to neither D block. Hence each D block needs
block-alternating sum+1.

The remaining M-array has nonzero total and needs a group-M column,
which belongs to neither D block. With A<=3, outside-M columns can
only be an opposite-coefficient pair in one zero-total array. Since
each alternating has exactly one block owner, that pair cannot supply
+1 to both distinct D blocks. All-group-M columns cannot supply either.
Thus A>=4 and R>=29. This handles every C/X budget at Pi25.

### 5.2 Pi26

A<=2 forces at least one canceled M-array. Its D is1 in (M,U).
The second D must also be in M: an outside D would require its array
and the two unchanged M-arrays, giving A>=3. Call its coefficient b.
Since A>=1, C<=1 and the Z-free S rows exclude C1, so C0.

If b!=1/2, all four endpoints have nonzero Gamma and f>=9. The line
sum48 leaves only two excess units, so none can have a separator.
Equation T2 then forces b=1. The proved two-unit-D obstruction for
Pi<=26,A<=2,C0 excludes the case, including mixed slack.

If b=1/2, T2 requires NT_U=1/2. The baseline costs are9,9,8,8,5,5,
with four units of slack. Exactly one unit endpoint must be a separator
at f13, with all others at baseline. Its F5 separator has marginal1/2.
The other unit endpoint cannot receive a separating glue; the two
half-valued endpoints are glue-free. Thus its F5 line cannot be dense
and must be a single positive glue+1/2 to an inactive center x.
The inactive f5 line at x is then dense and forces nonzero glues to
the other two vertices on the separator's half. At least one is a
glue-free half-valued endpoint: after excluding the other unit endpoint
and x, only one other inactive vertex is available. Contradiction.

### 5.3 Pi27: extension of the two-unit-D obstruction

The only budget is A1,C=X0. Both D must be1 in M to cancel two arrays.
The only alternating belongs to the third M-array, so the D blocks
need actual star mass2 each. The line sum50 has four excess units
above active f9 and inactive f5.

An active endpoint can have a star with either D source only on a
separating line for its own source. Such a line forces f>=13: two
main lines cost8, any additional nonempty N/star line costs at least4,
and equality12 would be glue-free, contradicting Gamma1/2. A separator
without N instead costs at least5. Thus at most one active endpoint
can contribute, and if it does, it has f13 and all other centers
are at their minimum.

For that endpoint, T2 gives total separating marginal0. Two nonzero
separators would cost at least10 beyond the main cost8. A nonempty
zero-marginal line with a glue costs at least6, also too much. Hence
the relevant separator is glue-free, of cost4. The two main lines
are F4 and F5, with main marginal1/2. The F5 is the dense, star-free
single-N form. To have an actual separator star, a balanced+1 cancels
the N=-1 on the F4 main line, with its partner-1 on the separator.
The separator star is therefore+1. These conclusions follow from
the V cells and balanced parity; no surviving pair of main stars is
assumed. That one active center contributes at most1 to the two D
blocks, while the two inactive f5 centers contribute at most2. This
cannot provide their required total4.

Otherwise no active endpoint contributes to either D block. Every
inactive f<=9 has one nonzero-marginal line of marginal-1/2 and no
other nonempty line. Its star source is fixed, so each inactive center
can contribute to at most one D block. Both must therefore contribute
exactly2, one to each block. A star of coefficient2 on an N-free line
of marginal-1/2 requires its corresponding glue to equal3/2. Both
inactive centers thus have f>=7. The four excess units force both f7
and all active f9. But an f7 line with marginal-1/2 has glue form
(-1/2,t,-t); a glue3/2 forces another glue-3/2 at a distinct opposite
center. Only one opposite center is inactive; the other large glue
would hit an active f9 line, whose glue magnitudes are1/2. Contradiction.

Pi>=28 cannot fit A>=1. This exhausts the disjoint-source case.
