[petalas]

Summary

PR3547 left seventeen placements unresolved by its supplied maps. This package completes the coherent boundary-map classification for its thirty exact placements.

same exact marked pairs / terminal representatives
  29 placements -> coherent GF4 map -> every finite three-terminal graph
   1 placement  -> no coherent characteristic-two map in any dimension
  all 30 stars   -> GF4 transfer, using a hub-dependent map at the exception

Fifteen coherent placements are new beyond PR3557. The unique exception is marks E,F with ports E+G,F+G. Its separate star map changes unreplaced queries using the hub direction. The written impossibility excludes coherent maps over every characteristic-two field; arbitrary graph-dependent transfers at that exception remain open.

All changes are isolated in research/p112-two-port-transfer-classification/. Proof, exact table, source origins and replay. The actual-reader pair floor is preserved in the stated complete-plane model. No cheaper decoder, physical supplier, kappa gain or selected-frontier change is claimed.

Credit retains Philippe Van Velzen, ctoth, the actual-reader lineage and Nick Petalas's PR3506/3524/3535/3547/3557. Used proof/runtime inputs are canonical ancestor bytes with exact original provenance. Comparison-only pending PRs are separated in COMPARISONS.json. Scientific replay needs no foreign Git objects or network.

Evidence

• Before: the original fixed maps admit thirteen coherent placements; PR3557 repairs one. The initial hypothesis that all thirty admit coherent maps is refuted.
  After: two separate exact implementations agree on twenty-nine coherent maps, thirty stars, all 3,312 compared subsets, 2,060 source and 2,736 reader premises, thirteen obstruction predicates and the complete ordered stream.
• Twelve optimized real-CLI corruptions reject and fourteen initial/restored runs pass. The rational/GF4 elimination and integer-minor/all-annihilator programs do not import each other.
• At actual head f0cb59da21fbde87b4ca6a7480b0905d6f6fb9bc, CPython 3.12.15 registered checks, metadata/pins, committed-byte/source-history checks, focused Ruff/format, official export integrity and all three Git-free scientific/control commands pass. Fifteen source files and sixteen export files retain their bytes and identities after replay. Copies are owned regular files with distinct inodes; the shared runtime is unchanged.

TMPDIR=/your/private/tmp python3.12 -W error -B scripts/research_package.py --package research/p112-two-port-transfer-classification
python3.12 -W error -B scripts/check_research_commit.py --package research/p112-two-port-transfer-classification --check-source-history

Old fields have characteristic zero and terminal representatives are fixed. Other marked pairs/scalings, larger/repeated interfaces, physical execution/restoration, hosting, stock, precision, compiler/all-size admission and full supplier/multiplication obligations remain open. Independent human and formal proof-kernel review are unclaimed. The changed algebraic scope does not meet the original coverage-law card's full physical acceptance gate.

Merge Danger

Door: two-way

An isolated research package adds an opt-in registered replay. Frozen inputs and the selected result remain unchanged.

Blast Radius: research