[petalas]

Summary

Prove the exact shared-port graphic query transfer criterion and its scoped non-Fano application.

selected graph edges
  -> differences of terminal potentials within each component
  -> one additional Q_e - Q_f test for two old query ports
  -> transfer for every finite graph on those three fixed terminals

Of the 30 displayed rational/GF4 attachments, 13 preserve the supplied marked transfer and 17 fail it. The compatible cases preserve the literal pair-reader floor for arbitrary finite attached graphs. The failures concern these fixed maps. They do not prove that another transfer is impossible or that a decoder saves.

The old-field application is characteristic zero. A six-term actual rank-(4,4) failed-map control costs 14 against baseline 4. No supplier, new kappa or selected-frontier change is claimed.

The package retains all runtime and mathematical premises on canonical ancestry, with exact origins and predecessor credit. It builds on Philippe Van Velzen, ctoth and the actual-reader lineage through PR3414/3463/3472/3489/3491/3505, plus Nick Petalas's PR3506/3524/3535. PR3535 is comparison/handoff context; no foreign Git object is required for scientific replay.

Evidence

• Before: PR3535 leaves cross-port shared directions open. At marks E,F and ports G,F+G, an explicit subset recovers E over Q but fails under the retained GF4 map.
  After: The written component-flow/path proof gives the exact partition criterion. Separate rational/GF4 elimination and integer-minor/all-annihilator implementations agree on all 30 cases, 64 subsets each, the literal endpoint/ranks/bill and 1,024 five-vertex graph subsets.
• CPython 3.12 direct and registered commands pass at 760659d. Four mutations cause 8 rejecting optimized CLI runs, with 4 positive runs before/after restoration. The repinned-template fault reaches the scientific base predicate.
• Metadata, source pins, final-HEAD check_research_commit.py --check-source-history, focused Ruff and formatting pass. Official committed export creation/check and all 3 actual Git-free scientific/control commands pass with PATH=/nonexistent; source, export and shared-runtime identities remain unchanged.
• Replay: python3.12 -B scripts/research_package.py --package research/p112-multiport-query-boundary. Proof, source closure, exact result, controls, direct commands and limits are in that package. Independent human and formal proof-kernel review are unclaimed.

Physical acquisition, coefficient execution, dirty restoration, frames/hosting, precision, stock, compiler/all-size admission and a complete consumer remain separate. Other marked pairs, odd characteristics, larger interfaces and alternative transfer maps are not covered. No Lean, Docker or broad selected-supplier replay was run.

Merge Danger

Door: two-way

One isolated research package; frozen inputs, shared tooling and selected frontier are unchanged.

Blast Radius: research

Archival publication and passing source checks do not imply scientific acceptance.