[zxv]

The fixed helper680 split on PR5718 cannot be completed by ADD/SCALE programs at the three specified meetings, even with arbitrary rational coefficients and no program-length bound. This combines the constructive contact/retiming evidence with the stronger exact temporal obstruction. No new κ or universal P104 obstruction is claimed.

For cut encodings 1196/1199/1200, on both the original and validated retimed schedules, grant an arbitrary rational invertible map on all 8 or 9 first-meeting fields, arbitrary later linear maps and rational terminal normalization. Literal causal propagation and private input-coordinate probes force X + (ell*w - 1)L = 0, while an exact rank-two certificate makes X and L independent. Additional owned memory, different cut bases, extra contacts, changed native itineraries and different consumers are outside the result.

The retained construction supplies two useful positive results: its candidate generator discovers the published helper708 predecessor cut after another role's MOVE, and advancing MOVE16655 of role153 to just after16649 creates a legal contact with unchanged paid/scalar cost. Both are checked on all 753 arbitrary input columns, actual frames, lifetimes and targets. A broad tensor-span relaxation passes all six cases; the exact obstruction preserves the temporal coupling that relaxation loses. A singular fake fixes a main output while destroying an independent helper target.

Start with the combined publication overview, then the linked three-meeting proof. The earlier finite two-meeting searches remain supporting provenance, not the headline theorem; their overlapping counts are not added together. Frozen source programs, contact lists, receipts and attribution remain unchanged. Generated JSON accounts for most of the retained diff and is checked by the publication replay. Duplicate vendored predecessor files are deliberate immutable copies with exact byte joins.

The prospective 18(H4+H5-H9) target credit is unearned. The reconstructed source invoice remains 66,348 paid jobs and 12,976 productive fields; the selected frontier is unchanged.

Validation

The registered Python 3.12 normal and optimized publication replay checks all six exact obstructions, constructive positive and chronology/frame/independent-target controls, the complete invoice, and byte identity between the two retained packages. It regenerates the contact lists and legal-retiming controls in temporary copies and checks semantic equality of frozen outputs. Historical exhaustive coefficient searches are authenticated but not repeated in this publication pass; their separate full replay remains available. Standalone export and committed-byte results are recorded in the publication validation record. The optional full source-history preflight is unavailable in this shallow checkout. No full native/selected/Lean/Docker or new caller/compiler/outer admission is claimed.

python3.12 -B scripts/research_package.py --package research/p104-boundary-reuse

Significance for researchers

A legal new contact and a favorable information-span witness do not necessarily compose into a realizable temporal program. This result removes coefficient and length limits from the earlier failure within the exact three-meeting interface, while preserving a reusable candidate generator and legal retiming fixture. Longer scalar words at those meetings are no longer a justified next search.

Open next questions

A successor must change an actual contact/delivery condition, the cut basis, or an explicit owned-memory boundary that defeats the two-direction contradiction. Extract the new causal transfer first, then price the complete acquisition, transport, restoration and cleanup against the target allowance. No surviving changed construction or favorable complete invoice is supplied.

Credit flyingviper's PR5718, Ali Romman's PR5713 and PR5708, and the complete native, helper, controller, compiler and analytic lineage retained in the notices. Prepared for zxv with substantial OpenAI Codex assistance. Apache-2.0. These source-bound exact certificates are not independent human or formal-kernel acceptance of the multiplication theorem.