[CubeJerry]

Why this matters

The completed once-rounded native MOVE provider named the selected #6030
source. The QUERY-only precision work uses the different proposed
#6936 word. Shared frame labels or an abstract
ABI did not supply the missing source correspondence.

This package explicitly binds every actual raw #6936 MOVE to that provider,
after checking the operator, absolute phase, actual owner and complete source
lifetime. It leaves the selected catalogue and selected frontier unchanged.

Exact result

For all 6,323 raw MOVE occurrences and their 2,961 directed chart pairs,
authenticate the exact T_b T_a^-1 convention and emit a typed entrance/PAID/
exit triple carrying the word hash. The actual source's retained compiler
and native verifier are byte-identical to the provider's; all 1,095 frame
circuits are checked, followed by every chart's 18 Pauli images and complete
normalized 512-position separating column. Every factor is admitted by the
actual unchanged physical Recipe constructor.

An isolated discovery binding connects those recipes to the completed
rounded runtime. Only two declared imports are redirected; all numerical,
header, staging, cleanup and codec method bodies are unchanged. For positive
parent f<=cutoff and the full inherited geometry, the result is exactly
R_Delta((T_b T_a^-1)^(tensor f) x) on arbitrary signed public inputs, with
only true final-width overflow rejected before commit on the valid provider
domain. All9f axes belong to the containing actual-input error fibre.

The complete expanded invoice, both serial repetitions, source/template
work, fixed 4,368 roles for capacity64, full staging and historical backing
remain. No long-record/control absorption or rf<=h is inferred.

Pins, predecessor credit and novelty

Canonical base: 7c49fe32c55de02c2c873655c804b16e4c3fef0a.
Proposed source: 03626137d0bd57f09b0bf9597376183ce6fa9683, exact word blob
d93e67ca7628bf5531ae6e369da4dd7b7907b88f; later d360... changes only the
proposal's VERIFY timeout. The provider is the published
once-rounded selected-source package.

DreamingOfClouds' #6936 already checks finite native charts and credits
#6914 and #6930 for
owned rises and phase behavior. Those identities, the Pauli certificate
principle and the generic lift/round construction are credited prior work.
The new result is the actual source-to-complete-physical-interface implication,
including typed token isolation and exact recipe/owner/lifetime admission.

Verification

python3 -B scripts/research_package.py --package research/current-6936-rounded-move

Four normal/-O gates cover the complete finite family, all typed raw
occurrences, two source/lifetime scans and actual provider bindings at every
rank. Controls reject wrong word/hash/source/owner, Boolean aliases,
truncated/reversed chronology, wrong phase, input-Z and rank, public/owned
alias, missing CLEAR and a wrong current frame. Computational ASTs match
after removing only the two declared import redirects; the old selected
discovery/catalogue remains unchanged.

No numerical whole-MOVE execution is claimed. The new evidence is the
scoped universal source/interface proof plus finite source/certificate and
constructor/binding checks. Mandatory geometry still requires at least 4,096
records. No small positive child or factor-stub test is called a whole MOVE.

Limitations and κ status

No new κ; no unconditional multiplication theorem. All five global
assumptions remain at their full scopes. The new hash-namespaced token is an
explicit source-binding layer, not an installed enclosing compiler token,
expanded caller/header exporter, nested remap, scalar/LAMBDA service, helper
normalizer, main adapter or nonleaf rate. The proposal's changed alphabet and
owned-ascent policy are not adopted as global framework theorems here.
Actual placement, owner grants, global grid/radius and complete recursive
costs remain open; the valid-shape/disjointness/provider-return premises remain.

Prepared for @CubeJerry with substantial OpenAI assistance. Full predecessor
and native-source credit is retained. Separate model-assisted review is not
human or proof-kernel acceptance.