The Half-Scalar Does Not Yet Define an Integral Index
Premise superseded by Entry 1132. The general warning about rational frames is valid, but the particular half-scalar being audited was not a typed comparison map.
Sequence claim: seqclaim-d33957fc5dd1660cf20d7290.
Question
Does Entry 1128's rational comparison
prove that the physical soft-node lattice embeds with index two?
Audit
The node side is integral: the normalization--conductor difference row is primitive and has a primitive generator .
The target packet is different. The rank-twelve connection and (e_6) bridge are rational de Rham data. The existing filtered-cospan checker says explicitly that its displayed integral columns are independently rescaled integral representatives. It does not construct the Betti lattice in the (e_6) line.
Indeed a constant rational target gauge
preserves the rational differential module while changing the displayed scalar to (-1/(2q)). The choices (q=1,1/2,2) produce respectively
Thus the denominator two is not an integral invariant of the frozen packet.
Narrow result
Calling it index two would repeat the rational-frame error rejected in Entry 782. Fixing the index requires an independently derived integral Betti or relative-homology lattice, polarization, normalized period, or physical pairing. No new carrier datum is indicated.
Evidence and next falsifier
research/benincasa/checkers/rank12_soft_node_e6_integral_lattice_gate.py;research/benincasa/results/rank12-soft-node-e6-integral-lattice-gate.json;- Entries 627, 782, 1086, 1103, and 1127--1128.
The next admissible route is to normalize the (e_6) line from the original Cayley--Menger relative period: derive one source cycle and its intersection pairing with the (e_6) form, then compare it with the primitive nodal loop. Absent that datum, the rational comparison is the terminal statement.