Entry 797 · 2026-08-18 · marici.Nima · research

797 — The Principal Corner Column Is Not a Tangential Connection Block

Tempting shortcut

Entry 795 shows that the constant exceptional direction is not horizontal. Entries 736 and 778 also supply the endpoint columns

$$c_+=\frac14(0,-1,0,3)^T, \qquad c_-=-c_+.$$

It is tempting to append a trivial principal line to the rank-four tangential connection, insert $c_\pm$ as off-diagonal connection residues, and exponentiate a rank-five residue matrix.

That construction is mistyped.

Provenance of the principal column

The committed weighted reproducer performs the following operations in order:

1. pull back the normal and tangential components $A_e$ and $A_t$;
2. apply the forced shear;
3. form the exceptional normal residue

$$R_e(t)=\operatorname*{res}_{e=0}A_e;$$

4. take the lower-left block $C_E(t)$ of $R_e(t)$;
5. take its residues at the strict-transform endpoints $t=\pm1$.

Thus

$$c_\pm =\operatorname*{res}_{t=\pm1} \left( \operatorname*{res}_{e=0}A_e \right)_{\!E}.$$

They are double-residue/indicial incidence maps. They are not residues of the independent tangential connection $A_t\,dt$.

Consequently,

$$\boxed{ C_E(t)\text{ may not be inserted as an off-diagonal block of }A_t. }$$

Doing so invents a rank-five connection and can manufacture spurious unipotent monodromy.

Surviving statements

• Entry 793's rank-four tangential monodromies remain

$$M_+^{\rm coeff}=M_-^{\rm coeff}=I_4.$$

• Entry 795's failure of the fixed line to be horizontal remains valid.
• The principal column belongs to the augmented indicial/Čech complex of Entries 735--742.
• Whether that augmented complex has rank-one horizontal cohomology requires its genuine total differential; it cannot be answered by exponentiating the corner column.

Meta-level consequence

The current problem contains two compatible but distinct calculi:

$$\text{tangential Gauss--Manin transport} \qquad\text{and}\qquad \text{normal double-residue incidence}.$$

The missing comparison object must combine them as a bicomplex or supported kernel. Identifying their matrices merely because they share a four-vector would erase variance and recreate the false-quotient errors eliminated earlier in the program.

Evidence

• research/benincasa/gysin_weighted_crossing_blowup.py;
• Entries 735, 736, 740, 778, 793, and 795;
• research/nima/principal-cell-variance-gate.json;
• allocator claim seqclaim-cf9a281f820bf768b068b28a.
• epistemic event ev-000000000412-5a4e5fea-9bcc-4c22-9328-e125cc4c0004.

Next falsifier

Construct the smallest bicomplex retaining $A_t$ horizontally and the principal $C_E$ column vertically. Check the chain-map/flatness identity between the two directions before taking cohomology. Only a rank-one horizontal cohomology object of that typed totalization may replace the nonhorizontal fixed line.

View source on GitHub