## 1. Claim, scope and inputs

In the canonical rational PM2(6,2) model, a normalized identity with
matching corrections, R=28, exactly one N attachment vertex u, and at
least one nonzero normalized helper is impossible. All signs, balanced
N at u, mixed support X and idle loops are allowed. Here m=1 means
the N attachment count in the attachment partition; balanced N at other
vertices would give m>1 and is outside this claim.

The preceding lower-bound-31 component is the preliminary input.

The proved R27 exclusion is not being extrapolated. Section4 derives
every R28 support tuple, and sections5--6 prove the new tight cases.


Use the proved scale3, actual center star masses1, and

    R=Pi+A+C+X, h=d+z, 2(Pi-h)=sum f,
    F=g+2s+2n, Gamma(u)=sum helper-class values-1/2.

Matching rigidity O29, valid at R<=29, gives Omega=-2e_M and w=2M.
Normalization excludes D at m1. Every Z is at u in a distinct block
and a distinct nonzero N class. Each class has four possible N on
four distinct private lines. Its two h-side sums are -Z; two N have
coefficients -Z,-Z, one per side. A balanced class has even N support,
and a balanced pair has opposite coefficients on distinct lines.

The proved general equations are

    div(delta)_e=H0_e+sum_B beta_B*(1[e=U]+1[e=V]/2+1[e=W]/2)-w_e, (F)
    H0_ab=(B_ab+B_ba)/2,
    B_ab=-2*sum_(halves at a containing b) marginal,
    2 sum_B delta_B c_B=sum_B (Z_b-Z_a)(e_a-e_b),
    c_B=1_W-1_V.                                                (P)

For m1 each beta equals its Z. Mixed columns have zero contribution
to these equations and to V/h, center and block equations.

Proved local facts include:

- Inactive f>=5; strict excess is at least2. Inactive f<=9 has a single
  marginal -1/2 and every B entry is zero or one.
- Active f>=8, or f>=9 if Gamma!=0. Two nonzero classes require f>=13.
- Actual center mass requires a nonzero star. Nonzero Gamma requires
  a glue. Every N-containing line costs at least4, and F4 is glue-free.
- A one-N F5 line has marginal -n or -n/2; all its nonzero glues have
  magnitude equal to the absolute marginal. A one-N F6 has marginal0.
- Two opposite N on a line of cost<=7 give F4 (same cell, no glue or
  stars), F6 (marginal0), or F7 (marginal plus or minus their amplitude).
- Same-prefix classes from different groups occupy different V
  positions on each common line. Different prefixes share exactly one
  possible private line, the half containing u and both prefix mates.
- A nonzero class and a nonzero separating marginal require f>=13.
- Every Z block needs a loop; a Z-free flowing block needs at least
  two loops. A sole global Z costs C>=6. If its block has only one
  loop, the stronger argument gives C>=7 (proved Part G, L1+).

## 2. New universal single-Z, C6 source-flow lemma

**SZ6.** In any normalized identity with exactly one global Z and C=6,
coordinate flow divergence at its source edge is zero. D, N, correction
weights and mixed columns may be arbitrary. This lemma only uses (P)
and the proved loop costs.

Let the Z be a!=0 at u, source U={u,t}, in block B0. Its block needs
at least two loops, since one would force C>=7. The projected residual
rho-q_0*c_0 has support two if q_0=0, and six if q_0!=0, where
q_i=2delta_i/a and rho=e_t-e_u. One other four-supported block vector
cannot supply either residual. Thus at least two other flowing blocks
are needed, each Z-free and costing two loops. C6 forces precisely
these two blocks B1,B2, two loops in each, and two loops in B0.

Suppose B1 carries nonzero flow on U. Then U is one of its side pairs,
so c_1(u)=c_1(t)=epsilon in {+1,-1}, and q_1!=0. B0 has zero divergence
on U. From the two coordinates u,t of

    q_0*c_0+q_1*c_1+q_2*c_2=rho,

c_2 must have different values at u,t. If these are -1,+1, the equations
force q_1=0, contradiction. Otherwise one endpoint value is zero and
the other is plus or minus one. They force |q_2|=2 and |q_1|=1.

The source of B2 therefore contains one of u,t and an outside vertex x.
At x, c_2(x)=0 and c_0(x)=+/-1. The equation gives |q_0| either0 or1,
because c_1(x) is zero or +/-1. If q_0=0, the other three outside
coordinates with nonzero c_2 cannot cancel a term of magnitude2 using
q_1*c_1, whose magnitude is at most one. If |q_0|=1, cancellation at all
three of them requires c_1 to be nonzero at all three. But c_1 already
has two nonzero coordinates u,t, and has only four in total. Again a
contradiction. Therefore no flowing outside block has U as a side.
The source edge's flow divergence is zero, proving SZ6.

The exact checker tests all258 fixed normalized projection configurations
with a proposed source-side flow block. Their rational affine systems
are empty. Relabeling covers every source/block and both Z endpoints.
No three-block network support search is involved.

## 3. Three collision-safe local controls

### 3.1 A value2 single class cannot fit f<=13 on the half-grid

**Z13.** Suppose u has exactly one nonzero class of value2, all glues
at u are in {0,+1/2,-1/2}, and f(u)<=13. Balanced N may be present.
Then no local V/h/center solution exists.

Here Gamma=3/2, so at least three glues are necessary. Five N plus a
star would cost at least10+3+2=15. With three main N, there are three
distinct N-lines; total cost<=13 permits only F4,F4,F5. Its sole
marginal is +/-1/2 by the F5 glue-magnitude rule, contradicting Gamma.
Four main N require four distinct lines and cost at least16.

With precisely two main N=-2 on distinct lines, a nonzero marginal
on a main line costs at least7: an F5 marginal would be1 or2 and its
glues violate the half-grid, while F6 has marginal0. Two such lines
would exceed13. A nonzero N-free extra line must be F5 after the two
F4 main lines; its marginal has absolute value1/2, not3/2. The remaining
possibility is one main line with marginal3/2, necessarily three
glues +1/2. Its actual star vector is (1,-1,-1) up to permutation,
so that line costs11 and its partner at least4, again exceeding13.

The only other four-N possibility is two main N=-2 and one balanced
pair +t,-t on distinct lines. Its cost forces exactly three glues
+1/2 and one actual star of coefficient1. Effective stars l-r are in
[-1,0], so the aggregate N vector l-r-S is in [-2,0]. The positive
balanced N must collide with one main -2 at the same V cell. The
other main -2 must occupy the actual-star cell with effective value-1,
requiring two +1/2 glues on that main line. The remaining negative
balanced N is away from that star and requires t<=1. The first main
residual -2+t is also away from the star, forcing t>=1. Hence t=1,
and it too requires two +1/2 glues on its distinct main line. At least
four glues would be needed, contradiction.

### 3.2 Two nonzero classes at f13 cannot have value2

**T13.** At a center with two distinct nonzero classes and f=13,
neither class has value2. Balanced N are permitted but cannot occur
at this cost.

Six N plus a star cost at least14. Five N would distribute as3+2:
different prefixes need at least four lines costing16; equal prefixes
have at least three lines with distribution(2,2,1), costing at least
5+5+4=14 because shared N positions differ. Thus exactly four N remain,
two per class. No additional N-free line fits. Four distinct N-lines
cost16, so only one or two common lines need be considered.

With one common line, the other two lines cost at least4 each. At
cost13, either the common line is F4 and the other pair is F4+F5,
or the common line is F5 and both others F4. In the first, cancellation
forces opposite class values, total Gamma=-1/2, and the singleton F5
marginal is -1/2. Its class values have absolute value1/2 or1. In the
second, the common F5 forces equal values and has no actual star;
the singleton stars sum to one, forcing both values1/2.

Two common lines require equal prefixes and different N positions.
The proved F5/F6/F7 collision rules leave only F5+F8 or F6+F7 at
cost13. In F5+F8 the values are equal. F8 with no glue has two stars
and forces value1/2. For F8 with two glues and one star, call the equal
class value v. The F5 marginal is v, and center mass1 gives the F8
marginal r_8=v-1/2. If its star is at an N position, its zero glue is
at the third position and r_8=0, so v=1/2. If its star is at that third
position, its zero glue gives r_8+1=v+1/2=0, so v=-1/2. In F6+F7
the values are opposite;
F6 has marginal0, F7 has marginal plus or minus the class value,
and total marginal -1/2 forces absolute value1/2. No case has value2.

The checker independently imposes first class value2 and center mass1
on every rational F<=13 glue/star mask, for the two-common-line and
both one-common-line position types. All3417 systems with faithful
nonzero masks are impossible. No bound on the second coefficient is
imposed. This lemma does not assert that every f13 B row is zero/one:
equal negative class values can produce two negative marginals.

### 3.3 Opposite different-source classes at cost<=15

**O15.** Suppose the two Z are at u with different sources {u,t_1},
{u,t_2}, values a,-a, one source equal to the M-edge at u, every
inactive B row is zero/one, and
flow is confined to these two blocks. If f(u)<=15, there is no identity.

Write t_1=u^ast, the matching mate. Neither helper source can be a side
of either block, since both sources contain u. Thus (F) requires

    H0_(u,u^ast)=2-a,        H0_(u,t_2)=a.                         (E)

Gamma=-1/2. Five N require four lines and cost at least16; seven N
plus a glue and a star cost at least17. Six N therefore mean two main
N per class plus a balanced pair: other3+3 or4+2 main distributions
would need five lines costing20. Their cost15 forces one glue -1/2
and one actual star1. The active B row is zero/one, so (E) forces a=1
and both entries one. On the unique glueline the effective star vector is a permutation of (0,1/2,1/2). The actual star is either absent there or is a unit vector there. The required aggregate N therefore has exactly two half-integer coordinates. The four main N coefficients are integers. The sole balanced pair lies on distinct private lines and can occupy at most one cell of this glueline. One arbitrary rational entry cannot supply both half-integer coordinates. This contradiction permits the sole actual star at every position.


With four N there are exactly three N-lines, a common line and two
singleton lines, and no extra nonempty N-free line. The common line
costs at most7. If F7, its sole marginal is +/-a=-1/2; the zero/one
active B row forces a=1 via(E), a contradiction. If F6, the singleton
pair is F4+F5; its unique marginal-1/2 is not on the common line.
Equation(E) again forces a=1 and the unique half to contain both mates,
contradiction. If common F4, every singleton cost pair at total<=11
that can satisfy Gamma=-1/2 has a unique marginal-1/2 except F5+F5;
the same half argument excludes
those cases. In F5+F5, the marginals are a or a/2 and -a or -a/2.
Their sum-1/2 forces either a=1 with M marginal1/2, or a=-1 with M
marginal-1. The matching H0 is then at most0 or3/2 respectively, while
(E) demands1 or3. This exhausts O15.

## 4. Complete retained28 budgets

Every Z at u can change only its own group array. At most one of its
blocks is in M. The two other M arrays remain nonzero; each outside
Z also creates its own nonzero outside array. Consequently

    A>=z+1.

For z>=3, each class needs two N and an actual star is necessary.
The five inactive centers contribute25 to sum f, giving Pi>=14+3z.
Every Z block costs a loop, so

    R>=14+3z+(z+1)+z=15+5z>=30.

Only z1 or2 remain. For z1, Pi>=18, C>=6 and A>=2. A2 forces the
matching array to cancel, so Z=2 and the proved value2 bound gives
Pi>=19. The complete R28 tuples are

| Pi | A | C | X |
|---:|---:|---:|---:|
| 18 | 3 | 6 | 1 |
| 18 | 3 | 7 | 0 |
| 18 | 4 | 6 | 0 |
| 19 | 2 | 6 | 1 |
| 19 | 2 | 7 | 0 |
| 19 | 3 | 6 | 0 |
| 20 | 2 | 6 | 0 |

Outside M, A>=4, so only Pi18/A4/C6 occurs. A>=3 does not by itself
assert whether the M array cancels; every allowed actual support is
kept in this table.

For z2, the proved two-class f13 bound gives Pi>=21. A>=3 and C>=2
leave precisely these ten tuples:

| Pi | A | C | X |
|---:|---:|---:|---:|
| 21 | 3 | 2 | 2 |
| 21 | 3 | 3 | 1 |
| 21 | 3 | 4 | 0 |
| 21 | 4 | 2 | 1 |
| 21 | 4 | 3 | 0 |
| 21 | 5 | 2 | 0 |
| 22 | 3 | 2 | 1 |
| 22 | 3 | 3 | 0 |
| 22 | 4 | 2 | 0 |
| 23 | 3 | 2 | 0 |

A3 forces one Z in M with coefficient2. If neither Z is in M, A>=5.

## 5. Complete one-Z exclusion

When Pi<=19 and C6, the active f<=11. A nonzero separating marginal
would require f>=13, so every nonzero marginal is on a half containing
its source mate t. Hence B_ut=-2Gamma=1-2Z. Every inactive f<=8 and
its B row is zero/one. SZ6 gives zero flow at source U. Equation(F)
therefore implies

    (1-2Z+B_tu)/2=w_U-Z,
    B_tu=2w_U-1.

For matching corrections w_U is2 or0, requiring B_tu=3 or-1.
Both contradict inactive zero/one lightness. This excludes all four
C6 tuples with Pi<=19, including X1 and the outside-M helper.

At Pi19/A2/C7, the Z is2. Its active f is at least11 and at most11,
and all inactive centers have f5. Every crossed glue at u is therefore
+/-1/2 (the proved F5 magnitude rule at its opposite inactive line).
Z13 excludes it. This local contradiction has no flow or mixed premise.

At Pi18/A3/C7, the line sum is34. Inactive strict excess begins at2,
so the active f is9 and all other f are5. The proved f9 law forces
Z=1, one dense F5 main line with marginal+1/2 and glue vector
(-1/2,+1/2,+1/2), and one glue-free F4 line. In that private partition,
the positive glues force their opposite f5 columns to have all three
glues nonzero; hence both other row centers are selected there as well.
All five inactive marginals are-1/2. The three row marginals sum
+1/2-1/2-1/2=-1/2, but the three column marginals sum-3/2. They sum
the same glue matrix, contradiction. This treats the new C7 budget
using its explicit unchanged local hypotheses.

Finally Pi20/A2/C6 has Z=2. Every inactive f<=7. If the active center
has no nonzero separating marginal, SZ6 gives the same impossible
source equation B_tu=3. Otherwise f>=13; the line sum is38, so active
f13 and all inactive f5 are forced. Their crossing glues again force
the half-grid, and Z13 excludes it. All seven one-Z tuples are closed.

## 6. Complete two-Z exclusion

### 6.1 Consequences when C<=3

Only the two Z blocks can flow: an outside Z-free flowing block would
cost two loops in addition to their required two. Both c vectors vanish
at u. Projection(P) gives a+b=0 and Gamma=-1/2. Equal sources give
zero projected right side; their distinct nonparallel c vectors force
both flows zero, requiring two loops per Z block and C>=4. Thus the
sources differ whenever C<=3.

### 6.2 Pi21

The active f is13 and all inactive f are5. A3 implies matching Z=2,
contradicting T13, regardless of C or X. This includes the C4 cell,
where flow might otherwise leave the Z blocks.

For A>=4 the allowed tuples have C<=3. If either source is the M-edge
at u, O15 applies, including when that helper belongs to an outside-M
group. Otherwise both sources have correction weight0. The opposite-
class f13 subcase has common F4 and singleton F4+F5, so its unique
marginal is-1/2 and every B entry is zero or one. At those two sources,
equation(F) requires H0=-a and H0=+a. Their nonnegativity forces a=0,
contradiction. Thus every Pi21 tuple is closed. Helper-group membership
and matching-source membership are distinct throughout this step.

### 6.3 Pi22

Here A<=4, so one Z is in M; C<=3. The active f<=15 and every inactive
f<=7 by subtracting the active minimum13 and the other four inactive
minima20 from total line cost40. Their B rows are zero/one. O15 closes
all three tuples, including C3 and X1; no idle-loop assumption was made.

### 6.4 Pi23/A3/C2: the new largest private budget

The matching Z is a=2, the other b=-2, and their sources differ. The
line sum is42. The active f<=17 and every inactive f<=9, so all inactive
B rows are zero/one. If active f<=15, O15 applies. Active f16 would
leave inactive sum26, an excess of one over25, impossible. Thus only
f17 remains, forcing all inactive f5 and every active glue to +/-1/2.

Let N be the N count at u. Gamma=-1/2 requires a glue and center mass
requires a star, so N<=7. For every case below the active layer has
only one nonzero-marginal line, of marginal-1/2. Its B row is zero/one,
and (E) would require H0_(u,t_2)=2, though H0<=1.

- **N4:** There are three main N-lines; four would cost16 and leave
  only one unit for a glue, producing a lone F5 marginal of absolute
  value1 or2, not1/2. With three lines, each nonzero singleton marginal
  costs at least7 (F5 is incompatible with the half-grid, F6 is zero).
  A nonzero common marginal costs at least9: if the two N cancel it
  is an N-free marginal plus their cost4; otherwise both actual-star
  positions remain nonzero because their amplitudes are +/-2 and
  effective stars lie in[-1,1]. Two nonzero singleton lines would cost
  at least4+7+7=18; a common and singleton would cost at least9+7+4=20.
  An extra N-free nonzero line costs5, and cannot coexist with another
  nonzero marginal within17. At most one marginal is nonzero.
- **N5:** Main counts are3+2 and require at least four N-lines costing
  at least16. At17 there is room for only one line to have glue; every
  F4 is glue-free. No extra N-free line fits. The marginal is unique.
- **N6:** Main3+3 or4+2 would require five lines costing20. Hence there
  are two N per main class and one balanced pair. At cost17, either
  there is one glue and two stars (unique marginal), or three glues
  and one star. In the latter, their signs are two negative and one
  positive because Gamma=-1/2; the sole actual star equals1. Effective
  star entries lie in[-1/2,1], and aggregate N entries in[-3/2,1].
  Neither an isolated +2 nor-2 main N can fit. The two main N on the
  common line must cancel at the same cell, and the balanced pair
  must cancel or modify the two singleton main N at their cells.
  Their residual aggregates are k,-k on the singleton lines, and zero
  elsewhere. A line with aggregate zero and no star has no glue. If
  the star is on a zero-aggregate line, that line uses all three glues
  and is the only nonzero marginal. If it is on a singleton, the other
  star-free singleton either has marginal0 or needs three glues of
  form(-r,r,r). It therefore uses all three whenever nonzero. Again
  there cannot be two nonzero marginals. The exact collapsed-line
  control has11 permitted patterns, all with just one nonzero marginal.
- **N7:** Its cost14 plus a glue and an actual star exhausts17, so
  there is exactly one glue and one star; the marginal is unique.

This excludes the last two-Z tuple. No rational coefficient of a
three-N class or balanced pair was assumed to be an integer.
