Assume matching corrections, R<=29, m=2, d=0, z=2, distinct physical attachments, and one helper block in M and one outside M. Write their coefficients as a and b.

## 2. All cases with C<=3

No Z-free block can contain a nonzero loop: such a block requires at
least two loops, in addition to one in each Z block. Equal sources
are excluded by the two-block projection theorem; different sources
force |a|=|b| and |delta_M|=|delta_out|=|a|/2.

If A>=6 then Pi<=21. A canceling M-Z of coefficient 2 would already need Pi>=22,
so it cannot occur. Both active costs are at most ten and all inactive
costs are at most seven. No active separator is possible. On the M
source e, the outside group's source cannot equal e (equal sources)
or have e as a side (the antisymmetric projection obstruction).
Otherwise e is absent from that group. Its equation requires the
directed entry from the M source mate to its attachment to be 3.
An inactive mate has entry at most one; an active mate lies outside
its own source prefix and has entry at most two by the low-cost list.
This excludes every A>=6 case, including arbitrary extra zero arrays.

If A<=5, all alternatings lie in M and the outside helper array,
except possibly a=2,A5. Indeed nonzero arrays consume at least four
directions without cancellation and three with cancellation; any new
zero-total array needs two additional directions. In the exceptional
a=2,A5 case, Pi<=22, whereas |b|=2 gives two active costs at least
eleven and hence Pi>=23. That exception is impossible. Every other
case lies in one of the 70 source-array spaces proved in S10.4.

Thus C<=3 is completely excluded, including the new Pi23/Pi24 and
third-loop boundaries. No inactive cost-ten lightness is used.

## 3. Budgets with C>=4

Array counting and the single-class cost bounds leave only:

* A3,C4,Pi22, with a=2;
* A>=4,C4,Pi20 or21;
* A4,C5,Pi20.

Larger A/C values only reduce the possible Pi. At Pi20 both Z values
are 1/2 and both active centers are glue-free f8; every inactive is f5.
At Pi21 the active cost sum is at most eighteen and inactive costs
are at most seven. Every alternating stays within the four stated
arrays: no additional zero-total array fits. The possible A3 case
also has exactly its required arrays.

If all loops stay in the two Z blocks, the 70 source-array contradictions in S10.4 apply. Otherwise precisely one Z-free block has loops. It
uses at least two, so at C4 each Z block has exactly one L3; at C5
at least one Z block still has exactly one L3. The additional block
may be idle: the demand arguments below also exclude a required
nonzero demand when its divergence is zero. We do not remove its
columns from a literal-row relaxation merely because its flow is zero.

### C4: equal magnitudes and the low-cost split

For one L3 at each Z block, the projected contribution of each Z is
its coefficient times a vector with six entries +/-1. The residual
must equal a multiple of the third block vector, which has two zero
coordinates. At either zero coordinate cancellation forces |a|=|b|.

For A3, a=2 would give |b|=2 and Pi>=23, contrary to Pi22.
For A>=4, Pi<=21. The common magnitude is therefore 1/2 or1.
Magnitude one forces both coefficients +1, both active costs nine,
and all inactive costs five. If the extra block has nonzero flow,
the 84 systems in S10.4 apply: 68 have exact affine separators, while the other sixteen contradict the forced f9,f9,f5^4 directed-entry conditions. The sixteen are not claimed to be affinely inconsistent; their exclusion uses the support costs. Same-block residuals cannot be supplied by a distinct third block.
If the extra block is idle, its divergence is zero. The M block
contributes net one at its source and one M-side. If the outside
block covers the remaining M-edge, it cannot also cover the M source,
whose H0 is at most 1/2; that source then has weight at most 3/2.
If it does not cover the remaining M-edge, that edge has H0 at most
one and no helper contribution. Either way F fails. This demand argument preserves every idle-loop column.

For magnitude 1/2, at most one coefficient can be negative. If both
are positive, both active B rows vanish and H0<=1 everywhere. Each
one-loop Z contributes net +1/2 at its source and one side. The M
block contributes total one on M; the outside block contributes at
most 1/2 there. Each M-node receives at most one. Thus

    D_e=H0_e+net_Z(e)-2,  e in M,

is nonpositive at every M-node and strictly negative at at least two:
equality would require net_Z>=1, possible at at most one node given
total net_Z<=3/2. One additional flow has only one negative divergence
node and cannot supply these demands.

If one Z is negative, its center has cost ten and two main marginals
-1/2,-1/2. Its directed source entry is two, but every outside-prefix
entry is at most one. The other center is glue-free f8 and every
inactive is f5. Thus H0<=1 except possibly at the negative Z source,
where H0<=3/2. At that source the negative net helper contribution
-1/2 lowers the total to at most one; every other negative-Z net
contribution is nonpositive. The positive Z adds at most 1/2 at an
M-node. Therefore D_e is strictly negative at all three M-nodes,
again impossible for one additional flow.

### C5: the remaining arbitrary flow value

Here Pi20, so a=b=1/2, both active B rows vanish, all inactive rows
are zero/one, and H0<=1. There is only one Z-free flow block.

If both Z blocks have one L3, the preceding nonpositive-demand proof
still applies; using three loops for the Z-free block does not give
it a second negative divergence node.

If the outside Z has two loops and the M Z one, its flow value may
be arbitrary. Its group nevertheless contains at most one M-edge.
The other two M-edges receive no outside contribution, and the
one-loop M Z contributes at most 1/2 to either. Both therefore have
strictly negative demand. One extra flow cannot supply both.

If the M Z has two loops and the outside Z one, the M source has
H0<=1/2 and receives 1/2 from its own helper and at most 1/2 from
the outside block. Its demand is strictly negative. On the other
two M-edges the M block's net contributions sum to 1/2, independent
of its unrestricted flow value. Their H0 values sum to at most two,
and the outside block adds at most 1/2 in total. Their combined
demand is at most 2+1/2+1/2-4=-1, so at least one of them also has
negative demand. The single extra flow cannot supply two negatives.

This exhausts C5 without a coefficient restriction on the two-loop
block. Hence every one-in-M, one-outside placement is excluded.
