## 1. Notation and stated scopes

Normalize the single attachment to u=0, with matching
M={01,23,45}. Matching rigidity O29 is already valid at R≤29, giving
Omega=-2e_M and w=2M. Write

    R=Pi+A+C+X, h=d+z, f(x)=g(x)+2s(x)+2N(x),
    2(Pi-h)=sum_x f(x), Gamma(u)=sum Z-1/2.

Here s counts actual stars, not effective stars. At one attachment d=0.
Each Z occupies a distinct block and a distinct nonzero N class; each class
has two h sides whose N sums are -Z. The minimal main pair has coefficient
-Z on each of two distinct private lines. A balanced pair has opposite
coefficients on distinct private lines. Different source prefixes share
only the half containing u and both source mates. Equal prefixes from
different groups have different V positions on every common line.

For a private line with glue vector l, marginal r=sum(l), and aggregate N
vector nu, actual stars are

    S=l-r*1-nu.

Inactive centers have f≥5, no f6, and their strict excess over5 starts at2.
An inactive f≤9 center has one marginal -1/2 and B entries0 or1. Its f5
line is either sparse with one glue -1/2, or dense with one positive +1/2
and two negative -1/2 glues. All its actual stars are nonnegative.

Use the proved necessary equations

    H0_uv=(B_uv+B_vu)/2,
    B_uv=-2 sum_(u-halves containing v) r,
    div(delta)_e=H0_e+sum_B beta_B*(1_U+1_V/2+1_W/2)-w_e,
    2 sum_B delta_B c_B=sum_Z Z*(e_mate-e_attachment),
    c_B=1_W-1_V,
    actual-star sum in B + alternating sum in B = beta_B.

Mixed columns vanish in all these equations and in V/h. At m1 beta is the
Z coefficient. Every Z block needs a loop; a Z-free flowing block needs
two loops. A sole global Z requires C≥6.

The **SZ6** lemma is used only at C=6: a sole global Z has zero
flow divergence at its source. It is not extended to C7 or C8. The proved
two-class **T13** lemma is used only to exclude a value2 class at f≤13.
The old O15 N-placement shortcut is not used; the all-star-placement parity
argument is used explicitly in section5.

## 2. Two additional glue obstructions

### 2.1 Five inactive f5 centers permit at most one positive active glue

Every nonzero active glue has magnitude1/2, forced by its opposite inactive
f5 line. A positive active glue forces that opposite line to be dense,
so the opposite center and the other two centers on the active half all
have their unique glue lines in that private partition.

Two such positive glues in different partitions would force three inactive
centers in each partition. The two three-element sets overlap among the
five inactive centers, forcing a second private line at an f5 center.

For two positive glues in the same active row, both opposite columns are
dense. The two other rows each receive two negative -1/2 glues, hence are
dense too and have positive +1/2 glues in the third column. That column
would have marginal 1 plus its active entry, one of1/2,1,3/2, rather than
the required -1/2. Thus there is at most one positive active glue.

Consequences used below:

- With a sole Z whose Gamma=Z-1/2 is positive, there is exactly one active
  +1/2 glue and no negative glue, so Z=1.
- At any active center the sum of separating marginals is at most1/2.
- If two Z have opposite values, Gamma=-1/2, so there are either one
  negative glue, or two negative and one positive glue. In particular
  B_uv≤2 for every v.

The checker confirms all435 pairs of potential positive active glues:
405 different-partition conflicts and30 same-partition conflicts.

### 2.2 One exceptional inactive f7 center is impossible

If one inactive center is f7 and every other inactive center is f5, the
f7 line has glue vector(-1/2,t,-t), up to permutation, with t nonzero and
|t|≠1/2. Its two unusual glues cross two distinct centers. Only u could
be an unusual crossing partner, since every other inactive glue has
magnitude1/2. This is impossible, regardless of the active coefficient
or balanced N. No second or zero-marginal line fits at f7.

## 3. Complete helper and budget cover

At most one Z block can lie in M. Its two other arrays remain nonzero;
each outside Z creates its own nonzero array. Hence A≥z+1. For z≥3,
the five inactive minima, two N per class, and an actual star give

    sum f≥25+4z+2, Pi≥ceil((27+4z)/2)+z=14+3z,
    C≥z, R≥15+5z≥30.

Only z1 and z2 remain. The necessary arithmetic envelope
contains40 m1 records,20 with z1 and20 with z2. Some are deliberately
overinclusive: for example Pi18,A2 forces a sole matching Z of coefficient 2, whose minimum
active cost11 already gives Pi≥19. Every actual case is inside the following
location refinements, with every C/X split retained:

| Helpers | Largest possible Pi | A minimum | C minimum |
|---|---:|---:|---:|
|One Z in M, value2|21|2|6|
|One Z in M, value not2|20|3|6|
|One Z outside M|19|4|6|
|Two Z, one in M with value2|24|3|2|
|Two Z, one in M with value not2|23|4|2|
|Two Z outside M|22|5|2|

Earlier R≤28 cases remain covered by their proof. The argument
below closes all40 necessary R29 records directly.

## 4. One Z: closure including the C7 boundary

Let a be the Z value, t its source mate, and T the sum of separating
marginals at u. Then

    B_ut=1-2a+2T.

At C6, SZ6 and the source equation give, when the inactive source mate is
light,

    T=(3-B_tu)/2 in {1,3/2}      if the source is an M-edge,
    T=-(1+B_tu)/2               otherwise.

With no separator the required inactive entry is3 or-1, contradicting
lightness. A nonzero separator requires active f≥13.

### Pi18 and all value2 cases

At Pi18 the line sum is34. The active cost is9 and every inactive cost5;
active f8 would leave the impossible inactive excess1. The stated f9
law gives a=1 and a dense main line with two positive glues, contradicting
section2.1. This excludes every Pi18 tuple, independently of C or X.

For a=2, active f≥11. At Pi19 active f11 and all inactive f5 are forced;
Gamma=3/2 contradicts section2.1. At Pi20, the alternatives are active
f11 with one inactive f7, active f12 with excess1, or active f13 with all
inactive f5. Section2.2, parity, or section2.1 excludes each.

At Pi21, A2 and C6 are forced. Active f11 or f12 has no separator and
every inactive center is light, so SZ6 excludes it. Active f13 would leave
a sole inactive f7; f14 would leave excess1; f15 leaves all inactive f5
and again contradicts Gamma=3/2. No C7/C8 source-divergence claim is used.

### Remaining Pi19 cases

At C6, active f≤11 and inactive f≤8; no separator fits, so SZ6 excludes
the case. At C7 only the noncanceling matching-Z budget Pi19,A3,C7,X0
remains. The line sum36 has these alternatives:

- Active f8 is the glue-free half class, leaving one inactive f8. Its
  two-glue vector(+1/2,-1) requires an unusual -1 glue to u, impossible.
- Active f9 leaves one inactive f7, excluded by section2.2.
- Active f10 leaves excess1, impossible.
- Active f11 leaves five inactive f5 centers.

For the last alternative g is odd and at most5. At most one glue is
positive. Three main N already use three lines of cost at least12;
four main N use four. Thus the only N possibilities are the minimal pair
or that pair plus one balanced pair.

With g1 the only nonzero class value is a=1. With the minimal pair there
are three actual stars: on the positive-glue main line they are +1/2 and
-1/2, and on the other main line +1. With four N there is only the actual
star1. With g3 the only negative nonzero value is a=-1 and every glue is
negative. A line with three negative glues leaves at least two actual
stars after its single N=+1, and the other main line another star; split
glues require still more. This exceeds the available two-star budget.
With g5, a=-2 cannot fit a main +2 in an effective entry of size at most
3/2. For a=-1 only one actual star1 is available: a main line carrying
that star and its N=+1 would require an integer-magnitude glue, whereas
both star-free main lines would require positive glues. Neither is possible.

The remaining a=1,g1 cases have seven or nine distinct blocks charged by
positive inactive stars. There are exactly six inactive glue completions:
four have seven charged blocks and two have nine. The one possible
negative active star can erase at most one of these blocks; the sole
helper can cancel at most one more demand. Hence A≥5, contrary to A3.
The four-N sole-star case has no negative active star and is stronger.
This closes C7 without using SZ6 there.

### Noncanceling Pi20,A3,C6

The line sum38 gives the following exhaustive active costs8–13.

- f8: only an inactive f10 could supply the exceptional source entry3.
  Its two F5 lines would have marginals -3/2 and1, needing glue magnitudes
  3/2 and1. Their crossing partners are inactive f5 or glue-free u; neither
  is possible. Every other inactive source mate is light.
- f9: the inactive excess4 is one f9 or two f7 centers. Its active glues
  have magnitude1/2. A sole f9 line is thereby forced onto the half grid
  and degenerates to F5. Each of two f7 centers needs two unusual partners
  and has only the other f7 as a possible partner. Both alternatives fail.
- f10: one inactive f8 is forced. The exact one-class f10 list is a=1/2,
  a=-1/2, or a=-1. The first two have an even glue count and integer
  Gamma; one glue -1 plus the other half-magnitude glues gives a
  half-integer Gamma, impossible. For a=-1 the main lines have marginals
  -1 and-1/2. Their glue vectors are a single -1 and a dense half-grid
  vector(+1/2,-1/2,-1/2). The positive dense glue forces all five inactive
  glue lines into its partition, while the f8 partner of the -1 glue is
  already selected in the other main partition. Contradiction.
- f11 leaves one f7, excluded by section2.2; f12 leaves excess1.
- f13 leaves all inactive f5. Section2.1 gives T≤1/2, whereas SZ6 needs
  T≥1 on the matching source.

For the f10 list: three main N cost at least12; four N at f10 force no
glue and a=1/2. With two N, F4+F6 gives a=1/2; F5+F5 gives a=-1/2 for
two sparse lines or a=-1 for one sparse and one dense line. This retains
the negative value-1 case, which cannot be omitted from the cost cover.

All one-Z budgets are excluded.

## 5. Two Z with C≤3: a short signed matching-edge obstruction

Let the Z values be a,b. No Z-free block can flow: it would need two loops
in addition to the two Z blocks. Projection at u gives a+b=0.

Equal source prefixes have zero projected right side. Their distinct
nonparallel block vectors force both flows zero, which needs two loops
per Z block, hence C≥4. For different source mates t_1,t_2, each other's
mate coordinate in projection gives

    2delta_1/a in {+1,-1}, 2delta_2/a in {+1,-1}.

This is valid even if an idle loop contributes to C3. Consequently the
effective side charges beta/2±delta of each Z block are precisely0 and
its signed beta. There is just one block with a positive beta, and it
charges just one side pair positively. Neither helper source can be23 or45.
At least one of the two inactive M-edges therefore has no positive
effective side charge. Its flow/source equation requires H0≥2.

Whenever every inactive center is light, this contradicts H0≤1 on both
inactive M-edges. For Pi≤23 lightness follows from active f≥13:

    f(inactive)≤2(Pi-2)-13-4*5≤9.

The sole larger tuple is Pi24,A3,C2,X0. A3 forces the matching Z of coefficient 2 and,
by projection, the other Z=-2 on a different source. Such a center needs
f≥15. For f≤14 five main N already use four lines; six N plus a required
glue and star cost at least15. Four main N on three lines are excluded
by8088 unrestricted rational glue/star masks, covering coincident and
different positions on the common line. No extra N-free line fits.
Thus the inactive bound becomes44-15-20=9, and the same signed side
obstruction applies.

The checker reads all105 class pairs at u:15 equal-source pairs,60
inconsistent different-source projections, and30 consistent projections.
Both amplitude signs give60 exact inactive-M-edge contradictions. This
uses no O15 placement assumption and no positive-helper premise.

## 6. The two new C4 cases

At C≥4 the only necessary records are

    Pi21,A3,C4,X1; Pi21,A3,C5,X0;
    Pi21,A4,C4,X0; Pi22,A3,C4,X0.

In the first two, active f13 and matching Z of coefficient 2 contradict T13,
regardless of where the remaining flow travels. The last two need new
block checks; no three-block or source-divergence lemma is assumed.

### Pi21,A4,C4

Active f13 and five inactive f5 centers are forced. There are four main
N. Five N at equal prefixes cost at least14, and at different prefixes
at least16; six N plus a star cost at least14. Four lines cost at least16,
so the main support has either two common lines or one common line and
two singleton lines. The active glue count is1 or3, with at most one
positive glue by section2.1.

The checker covers each of these three local incidence types. It allows
both helper values to be unrestricted rationals. Center mass gives a+b
from the specified glues; actual zero-star equations determine the finite
coefficient roots. Generic free coefficient families are also checked,
and none meets this cost. There are15 local forms (8,6,1 by incidence).

Embed these forms in every literal minimal main-pair placement with one
block in M. For each embedding, complete each inactive line by one of its
six proved F5 forms. Nonzero entries force its partition, and every other
line at that center is zero. This gives7568 completions of analytically
forced minimum private cores, not a census of arbitrary PM2 supports.
All pass the180 V and180 h literal equations and all six center masses.
Their block demands require **at least five alternating directions**.
Hence no A4 completion exists, regardless of flow or mixed columns.

### Pi22,A3,C4

One matching Z equals2. Active f13 is excluded by T13; f14 leaves the
impossible inactive excess1. Thus active f15 and five inactive f5 are forced.

For five main N with different prefixes, four lines already cost16.
With equal prefixes there are two double-N lines and one singleton line,
with minimum costs5,5,4. The g1/two-star alternative cannot fit both
double-N lines; one would be glue-free and cost8, giving at least17.
The g3/sole-star alternative has an isolated main N=-2 at some V cell
(three main N in a class have a singleton h side). Equal-prefix classes
have different positions, so that cell cannot be canceled by the other
class. But at most one positive glue gives every effective entry≥-1/2;
subtracting the sole actual star1 leaves aggregate N≥-3/2. Contradiction.

Six main N with equal prefixes need three double-N lines of minimum
cost5 each and leave no actual star at cost15; four lines cost too much.
Different prefixes need at least five lines. The remaining six-N option
is a minimal main pair for each class plus one balanced pair. Cost15
forces one glue and one actual star1. The glue is +/-1/2, so the other
helper is-2 or-1 and all four main N are integers. The two half-integer
aggregate entries on the sole glueline cannot be supplied by a balanced
pair on distinct lines. This argument permits the actual star anywhere.
Seven N plus the required star already exceed15.

It remains four main N, with active g1,3,or5. For the same three local
incidence types the fixed-value2 local calculation gives six forms, all
with the other coefficient-2 and a common line on which the main Ns
cancel in the same cell. Completing the five inactive F5 centers gives
7136 forced cores. Every core passes V/h and center equations; its B
demands require **at least eight alternating directions**, far above A3.
No flow geometry is used, so an additional flowing block does not help.

This excludes every two-Z record.
