[thu-nmrc]

PR #5551 supplies a complete conditional P104 candidate at κ = 0.00304153, but its scientific admission still needs a source-specific error/range application. This supplement makes the owned-COPY implementation explicit and supplies exact all-cut certificates and finite-format bounds for that actual word and its complete 265-axis caller. It claims no new κ and does not treat #5551's archive merge as scientific acceptance.

• Realize the eight owned COPYs as ADD into their initialized zero. A retained literal word and fresh source replay preserve every exact endpoint, MOVE request, paid/helper histogram and the 217428 scalar allowance, while preserving arbitrary injected owned residuals. Assignment-COPY would erase a real nonzero suffix response.
• Two independent exhaustive dyadic/Fraction interpreters scan all 25,311 cuts on 735 arbitrary public inputs and 743 state positions. The exact maximum prefix and final-suffix row norms are 131072 and 135415/4, with full equality witnesses. An independent induction/fixture bounds every scalar interval by 2^296, with denominator exponent at most 54.
• The credited active-fiber method gives a complete-caller interval bound 2^5844 in maximum-individual-field Hilbert norm, and 2^(5844+ceil((265f+1)/2)) in coordinate norm. Every active coordinate is retained; external spectators and the number of independent fields introduce no extra conversion factor in this norm.
• A fresh whole-caller census retains all 12,814 productive fields, reserves 248 owned query/local positions, and supplies finite ordinary recipe, integer range and storage bounds. Ordinary schema coefficients are distinguished from actual tape steps. Literal butterfly witnesses are distinguished from priced recursive services.

The original five native/ownership/compiler/precision-tape/analytic premises remain. This is a finite-template proof/application and independent author-side algorithm review, not a materialized recursive child backend, fixed-tape multiplier, formal proof-kernel certificate or selected-frontier promotion.

Validation: Python 3.12 stdlib; direct isolated normal/-O replay, both registered commands, committed bytes/source history and official closed export/check pass. A relocated export with no Git, empty PATH and read-only files passes both modes with identical stdout and unchanged inventory/hashes. See research/p104-cut-stability/VALIDATION.md, PROOF.md, ASSEMBLED-REVIEW.md and the retained independent capsules. Archived publication head: 3b154c52cb05922fc10eb8ce656a2a314bbb8187.

python3 -I -B research/p104-cut-stability/verify.py
python3 -I -O -B research/p104-cut-stability/verify.py
python3 -B scripts/research_package.py --package research/p104-cut-stability
python3 -B scripts/check_research_commit.py --package research/p104-cut-stability --check-source-history

Credit: preserve Ali Romman and the full P104 geometry/native/controller lineage. The active-fiber and prior grid/error methods retain Ross Richardson / Adam Gall and predecessor attribution; their older partial-word constants are not reused. Prepared for thu-nmrc with substantial OpenAI Codex assistance, including separate author-side subagents. New work Apache-2.0; retained notices unchanged.

Publication update (2026-10-26): this PR was archive-merged at head 3b154c52cb05922fc10eb8ce656a2a314bbb8187 by merge commit 129a0465b7459a9c673ccc2da8533b496913b932. The first hosted plan failed on an inherited standalone-manifest declaration before P104 replay. Later personal-branch metadata/validation commits were pushed after the merge and are not included in this PR's merged head. A standalone-preserving compatibility correction is being handled separately.

The current selected conditional frontier is PR5599 at κ=0.00310045. Its maintainer contribution map explicitly credits #5551's coupled-contact improvement and says #5557's stability analysis informed review. This is adoption of the #5551 mechanism in the new construction; #5557's sharper numerical constants are not transferred to the changed word. See the selected contribution map and the source-specific application review.