Rule out identity from nested P104 transport refinements

[Tairitsu-Aya]

The P104 identity transfer remains impossible after arbitrary nested subdivision of every original transport interval, arbitrary interleaving and common-chart linear resynthesis. The source-to-target rank is at most 56 arrays: the 56 sources outside the span of the eight private-drop survivors can contribute at most eight arrays; the other 48 sources account for the remaining upper bound. Identity requires 104. This upper bound is not claimed sharp.

This closes a specific part of the changed-intermediate-frame handoff in #2113. Its fixed-path rank-eight bound does not already cover the newly admitted intermediate charts. The result also differs from #2127's added-storage theorem with retained unsplit transports: here the old occurrences may themselves be replaced by nested chains. Routes outside their original intervals, extra cross-chart transfers, branching/fusion and changed endpoint obligations remain outside the theorem. No native correction cost, supplier admission or new kappa is claimed.

Only research/p104-nested-refinement-cut/ is added. The proof colors a proper chart target-side when it lies in the survivor span or contains at least two survivor lines. All 1,402 upward intervals avoid a crossing under every monotone refinement; each of the eight downward lifetimes crosses once and carries one array. A transcript argument permits arbitrary common-chart mixing, feedback and fresh source-independent storage without multiplying the counted crossings. The README gives the model, proof, counterexamples, attribution and concrete next steps.

Validation:

• Two code-disjoint implementations authenticate the same source and compare the entire receipt. One uses reduced binary bases and four sufficient interval cases; the other uses literal XOR spans and a separate constructive decision for any possible upward bypass.
• Both independently exclude all 529,984 direct X/Y common-chart slot pairs, before timing constraints. The second decision is also checked against all 892 strict intervals in two complete 67-chart lattices, including the full-chart boundary.
• Actual excluded new routes bypass the separator; dependent survivors, a ninth descent, nonnested/full-to-line replacements, equivalent bases and equal-dimension countercontrols behave as stated.
• Windows CPython 3.12.9 and registered Ubuntu 24.04 / CPython 3.12.3 replay pass. Linux: 1.52 seconds, peak RSS 53,752 KiB, one CPU.
• Twelve isolated export checks cover normal/optimized replay, unchanged files, missing/corrupt input, count-preserving receipt corruption and exclusive output. Metadata/source pins, selected-record integrity and whitespace checks pass.

python3.12 scripts/research_package.py --package research/p104-nested-refinement-cut

The scalar coefficients, original 280-column identity, native compiler, global hosting, precision and all-size/outer proof are inherited and are not replayed by this package. The generic transcript implication is written mathematics, not a proof-assistant certificate. Both independent implementations were prepared by the same author/assistant; no independent human review is claimed.

Credit Rohan Arun/#1972, Chafik Boukhalfa/#1991, Adam Gall/#1940/2090, Josie Elliston/#1071, Ali Romman/#1125, David Arias/#2082, Jeremy Diviney/#2113 and Ross Richardson/#2127. The earlier fixed-alphabet corroboration remains on a public fork branch without a competing PR. Prepared for Tairitsu-Aya with substantial OpenAI Codex assistance in mathematics, implementation, source reading and prose; Apache-2.0.

Task p104-side-correction, explicitly changed scope from a construction attempt to a scoped obstruction. Scientific base is canonical main 0fdd7d82de03e43c9e5a1000dfbcd380d1b687d2, priority revision 4. Canonical claim: #2113 (comment) . The selected frontier is unchanged. Exact task/overlap coverage and limitations are recorded in TASK.json.