[RossRichardson]

PR3019's matching passes its independent-contact support test but cannot implement the additive correction under the stated monotone frame contract. The relay 139 -> 91 -> 83 forces vector 113 into helper 91's frame before requiring that frame inside a plane excluding 113. At least 45 required coefficients at Y40 are therefore impossible, including the added coefficient C[40,1]=1/2.

Only helper 91 needs a consistent flag; the other 207 remain relaxed. All 9,765 choices reject: 477 miss 45 dependencies and 9,288 miss 46. A separate implementation reconstructs the exact local contact cut directly from frozen JSON, without importing the primary checker or predecessor code. Proof and scope, reproduction and credit, canonical investigation.

Local registered CPython 3.12 replay and optimized CPython 3.14 checks pass, including 8 behavioral groups, an actual nonnested relay that restores all 4,992 relaxed paths, source/receipt corruption and read-only Git-free exports. Exact committed bytes match all 11 package files and 3 external inputs. Hosted CI is separate.

This rejects one frozen matching for the correction increment I-G/2. It inherits the original channel's G/2 contract and proves no new lifetime minimum, supplier admission or kappa. Different matchings and changed access contracts remain open. Credit Nick Petalas/PR3019/3013/2996/2981/2920, Arham Tahir/PR2997 and the retained P104 geometry/native/compiler lineage. Ross Richardson with substantial OpenAI Codex assistance and two internal mathematical specialist implementations; Apache-2.0. No independent human or formal-kernel review claimed.