[CubeJerry]

Why this matters

The exact native direct MOVE still requires signed/divisibility admission,
and the existing total rounded positive child stops before the exit
monomial. Rounding there does not equal rounding the complete source map:
nearest-up rounding does not commute with negative phases, and an early
public-width downcast can reject an input whose complete final output fits.

This package supplies a local once-rounded complete raw MOVE for the
selected #6030 source. It does not bind
the distinct proposed #6936 word.

Exact result

For all actual selected raw entrance/PAID/exit triples, positive parent
f<=cutoff, arbitrary signed h-bit components and the full existing
nine-block geometry, pay a dyadic lift by g=rank*f and widen to
w=h+g+ceil((g+1)/2)+1. Execute the complete exact entrance, positive child
and exit, then perform one final nearest-up grid drop. The result equals
R_Delta(Ux) with the original absolute phase; on the stated valid provider
domain, only true final public-width overflow rejects before public commit.

A new source-consumer proof admits the widened record/row lengths at both
serial monomial factors. The positive child alone receives private
(f,cutoff)->(rank*f,rank*cutoff) transport. A containing actual-input error
fibre includes all 9f native axes. Under the explicit radius condition the
call is total with residual eta_X; no ideal/correlated-input premise is used.

Both serial repeated costs, complete expanded raw/coded fields, all template
and constructor work, inner and outer protected stages, metadata loops,
cleanup and retained history remain charged. The cap64 factory has 4,368
fixed roles. V'<=3V requires the separate condition rank*f<=h.

Novelty and source pins

Canonical base: 7c49fe32c55de02c2c873655c804b16e4c3fef0a.
The package authenticates 244 unchanged files and 109 shared identities from
the rounded positive leaf,
exact direct MOVE,
and retained private cutoff transport.
The generic lift/round idea and predecessor kernels are credited prior work;
the addition is their complete literal source/interface/lifecycle composition,
new serial width permission, and two actual-source rounding-order controls.

Verification

python3 -B scripts/research_package.py --package research/current-rounded-source-move

Four bounded normal/-O gates cover exact source joins, six actual metadata
consumer cases, invalid raw triples, and two independent original-circuit
algebraic controls at event 9896. One distinguishes the exact half-tie output;
the other has intermediate h=2 overflow while all 512 final rounded outputs fit.

Three structural instrumentation controls replace all three numerical
factor methods and deliberately use a tiny shape outside the nine-block
domain. They execute actual outer codecs and both protected coordinators,
checking success, named late rejection, final overflow, untouched public
input and retirement of all 4,368 roles. They are not numerical MOVE runs.

No new numerical whole-MOVE execution is claimed. Mandatory geometry
forces at least 4,096 records and the earlier exact predecessor took 815.504s
before widening, beyond this wave's 120s/no-repeat limit. A larger structural
attempt lost its tool session with no saved receipt; its exit status is
unobserved. The universal result is the explicit provider-contract proof;
the finite evidence has the limited scopes above.

Limitations and κ status

No new κ; no unconditional multiplication theorem. All five global
assumptions remain at their full scopes. Enclosing caller placement, the
actual global grid/radius/expanded cost bounds, nonleaf admission, flattened
compiler correctness, ownership composition and analytic recovery remain
separate obligations. No #6030/#6936 source correspondence is inferred.
The proof assumes valid shape/disjoint segments and retained provider
contracts; it excludes interruption and malformed late returns.

Prepared for @CubeJerry with substantial OpenAI assistance. Exact
predecessor/source/oracle credit and all original notices are retained.
Separate AI/model-assisted review is not human or proof-kernel admission.