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

## 1. stated inputs and complete budget cover

Use the lower bound 31 proof, Parts B,D,G,H, with its exact rational
row certificates. Matching rigidity gives Omega=-2e_M. Write
R=Pi+A+C+X, h=1, and sum f=2Pi-2. Source balance applies because flow
is zero and all nonzero helper classes have source U:

    NT_U = w_U-1 = 1 if U is in M, and -1 otherwise.             (T)

The proved private22 obstruction is local and independent of R27:
it gives Pi>=23. Its proof in Part H section8 uses only (T), endpoint
costs and the collision-safe F4/F5/F6 laws. The same proof gives an
endpoint cost sum at least22: a nonzero-Gamma separator costs at least13
and the other endpoint at least9; with Gamma0 it costs at least14 and
the other endpoint at least8.

An inactive center with f<=9 has one nonzero-marginal N-free line of
marginal -1/2, all its B entries are zero/one, and every actual star
has source equal to that half minus its center. A nonempty zero-marginal
line with a glue costs at least6. The F5 nonzero-marginal law says all
nonzero glues have magnitude equal to the marginal's magnitude, including
the proved scope with N collisions.

The complete coarse support cover is:

| D location/value | Possible Pi | Additional budget facts |
|---|---|---|
| Outside group M | 23,24 | A>=4; C=0 since C1 is impossible and C<=1. |
| In M, a!=1 | 23,24,25 | A>=3; Pi25 forces A3,C=X0. |
| In M, a=1 | 23,24,25,26 | A>=2; Pi26 forces A2,C=X0. |

C=1 is excluded by the Z-free S rows. All idle-loop budgets with C>=2
are retained until excluded below. Any larger Pi already exceeds28
with its required arrays. These rows include all mixed support slack.

The 403-column dual D21 is used only when C=X=0 and every alternating
belongs to group M; it then gives -D+t_U^M/2=-1=0. Its allowed columns
are all private columns, all N attached at U, all twelve group-M
alternatings and this one D. No extension to mixed columns, loops or
outside-M alternatings is asserted.

## 2. A useful unit-class endpoint bound

**Lemma U25.** If a=1 in M and flow is zero, then

    f(u)+f(v)>=25.

Suppose the sum is at most24. At least one endpoint has a nonzero
separating marginal. Both endpoints have Gamma=1/2 and f>=9, so at
most one can be separating, since each separator costs at least13.
Its separating sum is therefore NT=1. Cost13 would make its two
together main lines F4, with total marginal zero, and its separator
sum Gamma=1/2. Hence its cost is14 or15.

At such a center three main N-lines, or an additional nonempty together
line, would cost at least12 before the separator's cost at least5.
Thus there are exactly two main N=-1, on two together lines whose
combined cost is at most10. Their possible F4/F5/F6 marginals are all
nonnegative, including balanced collisions. But their required sum is
Gamma-NT=-1/2. Contradiction. This is the proved Part H section11
Pi23 argument with its purely local endpoint hypothesis stated explicitly.

## 3. D outside group M

Pi<=24 and the endpoint sum at least22 imply every inactive f<=9:

    f(x)<=2*24-2-22-3*5=9.

Thus all inactive B entries are zero or one. If U is an M-edge, the
two other M-edges are absent from the D group: a different perfect
matching containing U cannot contain one of them without being M.
Their F equations require H0=2, whereas H0<=1.

If U is not an M-edge, let e be the inactive M-edge. Unless it is a
side of the D group, F again requires H0_e=2. If it is a side, then
H0_e=2-a<=1 gives a>=1. The other side j joins the two inactive M-mates
of u,v and is not an M-edge. Its endpoints are light, so H0_j>=0,
whereas F requires H0_j=-a. Thus a<=0, a contradiction.

This excludes Pi23 and24, including X and every permitted alternating
count. It does not assume the signs of the inactive glues or of a.

## 4. In M, nonunit coefficient

At Pi25 the budget is A3,C=X0. All three nonzero M-arrays require a
column, so every alternating is in M. D21 applies and gives a contradiction.
It remains Pi23 or24; no restriction on their permitted C or X is
needed for the following private argument.

The endpoint sum at least22 gives inactive f<=9. At either inactive
M-edge F requires H0=2-a, while lightness gives H0 in {0,1/2,1}.
Since a!=1, necessarily a=3/2 or2. A center with this sole nonzero
class has f>=11: the proved f<=10 class-value list has no positive
values other than1/2 and1. A separating center has f>=13.

### 4.1 Pi23

Total cost44 forces endpoint costs13,11 and all inactive f5. The
separating endpoint alone supplies NT=1; at f13 its separating sum
is Gamma=a-1/2, forcing a=3/2. Its separating F5 line has glues of
magnitude one. Such a glue cannot cross the other active center:
there a nonzero separating line plus two main lines costs at least13,
and a nonempty zero-marginal separating line costs at least14, exceeding
f11. It cannot cross an inactive f5 line either, whose glues have
magnitude1/2. Contradiction.

### 4.2 Pi24

Total cost46 gives endpoint sum at least24. Either all inactive costs
are5 and the endpoint sum is26, or one inactive cost is7, the others5,
and the endpoint sum is24. There is no other inactive excess at most2.

If both endpoints have nonzero separators, they must both have cost13.
Their separating sums are both Gamma=a-1/2, totaling2 or3 instead of
NT=1. Hence there is only one separating endpoint, of cost13,14 or15.

If its cost is13, NT=1 forces a=3/2 and its F5 separator has all
nonzero glues of magnitude one. The other endpoint has cost at most13
and no nonzero separator; a zero-marginal separating glue line would
cost at least6 in addition to its two main lines, so cannot fit either.
Thus the separator's glues must cross inactive centers. All but at
most one are f5 and cannot take such a glue. The exceptional f7 line,
if present, is N-free with marginal -1/2 and glue form

    (-1/2, b, -b), with b nonzero and b!=+/-1/2,

up to permutation. A glue of magnitude one therefore forces another
glue of magnitude one to a distinct opposite center. Because the first
line separates u,v, this opposite half contains u but not v; its other
centers are inactive f5. None can supply that second glue. Contradiction.

If the separator costs14 or15, the other endpoint costs at most12 and
has no separating glue. No exceptional inactive f7 fits: these endpoint
costs would leave an inactive excess of at most one. Thus all separator
glues cross inactive f5 lines and belong to {0,+/-1/2}. Only one
nonzero-marginal separator can fit, since two cost at least10 before
the two main lines' cost8. Its required marginal is one.

On the forced half-grid, marginal one requires exactly two glues +1/2.
Its effective star vector has three nonzero entries (-1/2,-1/2,-1).
Each N can affect at most one star position, so actual-star count plus
N count on that line is at least three. Its cost is therefore at least
2+2*3=8, allowing arbitrary N collisions. But the available separator
cost is at most15-8=7. Contradiction.

This excludes all nonunit cases without extending D21 to their C/X
or outside-alternating slack.

## 5. In M, a=1: all but the four-alternating corner

U25 excludes Pi23. At Pi24 or25 it gives inactive

    f(x)<=2Pi-2-25-15 <=8.

Hence every inactive center is light. At an inactive M-edge, F gives
H0=1, forcing both directed B entries one. Each inactive center's
unique nonzero-marginal half contains its M-mate. It has no additional
nonempty line, since f<=8 and a nonempty zero-marginal line costs6.

No actual star belongs to any block with source U: active centers are
in U and cannot have source U, and an inactive center's star source
contains its inactive M-mate and so is not U. In particular the actual
star sum of B_U=(M,U) is zero. Its B equation therefore needs
block-alternating sum two.

If A<=3, every alternating belongs to group M: the two nonzero M-arrays
already need two columns, and an outside zero-total array would need
two more. Group-M alternatings are disjoint from the alternating
support of B_U. Contradiction, irrespective of the remaining idle loops
or mixed columns. This covers all Pi25 budgets and all Pi24 budgets
except Pi24,A4,C=X0.

Pi26 forces A2,C=X0, all alternatings in M, and D21 applies. Larger
Pi cannot fit. Thus the unique remaining support tuple is

    Pi24, A4, C=X0, D=1 in (M,U).                              (Q)

## 6. Equality facts at Q, including balanced N

At Q, U25 and inactive f>=5 force every inactive f5: each has f<=6,
and f6 is impossible. The two endpoint costs sum to26. They are either
both13 and separating, or the unique separating endpoint has cost
14,15,16 or17 with its mate of cost12,11,10 or9 respectively.

Costs14/15 are excluded by the U25 main-line marginal argument. A
value-one center cannot have f10 by the proved Part D section2 lemma,
so the16/10 pair is impossible. At17/9, the f9 mate has all glue on a
together line, and all inactive glues have magnitude1/2. The separator
at the f17 endpoint therefore has a forced half-grid and marginal one.
There is exactly one such line, costing at least8 as above. The two
main lines then have combined cost at most9 and nonnegative marginal;
no extra together line fits. They cannot have required sum -1/2.

Thus both endpoints have cost13 and separating marginal1/2. Each has
exactly two main together F4 lines and one separating F5 line. The
two main N are -1. Arbitrary balanced N reduce to exactly two possible
forms at each endpoint:

1. No balanced N. Both main lines have actual star+1. The separating
   line is N-free, has marginal1/2 and nonpositive actual stars.
2. One balanced pair. Its +1 cancels a main N=-1 in the same V cell,
   leaving that F4 line star-free; its -1 occurs on the separating F5
   line. The other main line has star+1. The separator has no actual
   stars and three glues (-1/2,1/2,1/2).

To justify exhaustiveness, six N plus a glue and an actual star already
cost15. Three main N would use three together lines before the separator
and cost at least17. Hence there are two or four N, with the extra two
a balanced pair. They cannot cancel both main N, since they have opposite
coefficients. A main F4 line containing two N must have zero actual
stars and zero glue, forcing its extra coefficient+1. Its partner on
the F5 separator is -1, giving the second form above. There is no room
for an additional nonempty line. This retains the known balanced sharing
phenomenon rather than assuming two main stars survive.

**Balanced bridge incidence.** If a balanced +1 cancels a main N in
one of these together V cells, its unique partner on the same h side
lies on the half consisting of the center and the other M-pair containing
that line's third vertex. Its V position is opposite the center's U-mate.
This is checked for all360 canonical N columns, by the literal V and
h-class incidences. Consequently, in form2 the negative separator glue
is the glue between the two active centers, whenever that separator is
forced by this canceled line.

## 7. The remaining alternating pair forces too many charged blocks

The two nonzero M-arrays each require a column. To give B_U alternating
sum two, at least one outside-M alternating is necessary. Its zero-total
array requires a second. At A4 there are therefore exactly two group-M
columns, each of coefficient -2, and two columns in one outside array,
with opposite coefficients. The four columns of an array have four
different block owners, a literal incidence fact. Exactly one of the
outside pair belongs to B_U, with coefficient +2; its companion is -2.

Every other alternating coefficient is -2. Thus each block other than
B_U, where beta=0, must have nonnegative actual star sum. The array
containing the selected B_U alternating has two possible types of other
block owner: another block (M,V) for V in M minus U, or a block whose
source consists of one active and one inactive vertex. This ownership
statement is checked on all45 choices of M,U and all four B_U columns.

If the companion belongs to (M,V), its B equation requires positive
star sum. But all possible stars with source V come from active
separating lines: inactive halfmates include their own M-mates, and
main together stars have an active mate in their source. Active separator
stars are nonpositive by section6. Contradiction.

Otherwise write the companion's source as {u,w}, with u active, w
inactive, v the other active center, and x the inactive M-mate of w.
Only two centers can supply stars in this block: v on a main together
line and x on its unique inactive line. Each contributes at most one.
Indeed v has at most one main line on half {u,v,w}, and all inactive
f5 star coefficients are1/2 or1. The companion alternating is -2, and
any additional alternating in this block is also negative. Its B equation
therefore requires at least two. Both stars must equal one, with no
additional alternating in that block.

The inactive x line has half {u,w,x}. Its star+1 is the dense N-free
f5 form, whose positive glue is opposite center v. This last position
is another literal ownership incidence of the same companion pair.
It activates v's separator on the complementary half consisting of v
and the remaining inactive M-pair V. Every separator star at v thus
has source V. No other center can contribute a positive star with that
source, whereas all block-star sums must be nonnegative. Consequently
the separator has no actual stars: v has form2 of section6.

The balanced bridge incidence makes its glue to u equal -1/2. This
activates u's separator on {u,w,x}. Its source is the inactive M-pair
{w,x}; the same nonnegative-block argument forces u to have form2 too.
Both active separator lines have negative glue to each other and two
positive glues to inactive centers. The four inactive f5 lines are
therefore all dense. After ordering active centers first, the entire
private matrix is forced to be

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

Each of the four inactive centers has one actual star+1. Their source
pairs are four different pairs, each containing one active vertex and
an inactive M-mate. Thus their block owners are distinct, and none is
B_U. All actual stars are now nonnegative: the active centers each
retain one main star+1 and have no separator stars.

The four positive inactive blocks each need an alternating column by
their beta0 B equation. B_U needs a further column because its star
sum is zero and beta=2. These are five distinct block supports, giving
A>=5, contrary to Q's A4. This count is solely a block-partition count;
no group-array count is added to overlapping block counts.
