P104 #5718 helper680 split cannot be repaired at the three tested meetings

research/p104-boundary-reuse/README.md

@@ -0,0 +1,40 @@
+# The tested P104 helper680 split needs more than these three meetings
+
+This combined contribution preserves a legal new contact on [PR5718](https://github.com/CrocSwap/integer-mult-bounds/pull/5718) and the exact obstruction that prevents three specified meetings from completing the proposed helper680 split. **No new κ or universal P104 obstruction is claimed.**
+
+## Main result
+
+For each of three fixed cut encodings (1196, 1199, 1200), on both the original and checked retimed schedules, no ADD/SCALE word at the three specified meetings can preserve the required output. The proof grants arbitrary rational invertible transforms on all 8 or 9 first-meeting fields, arbitrary linear later transforms, and rational terminal normalization. Exact causal propagation forces `X + (ell*w - 1)L = 0`, but the two first-meeting combinations X and L are independent. This removes the earlier coefficient-menu and program-length bounds **inside this specified interface**.
+
+Read the [complete scope and proof](../p104-three-meeting-controllability/README.md) first. Additional owned memory, different cut bases, extra meetings, changed native routes and other consumers remain outside the result. A tensor-span relaxation passes all six cases; the obstruction retains the temporal matrix coupling that this relaxation loses. A singular fake fixes one main output while destroying an independent helper target.
+
+## Constructive calibration and predecessor evidence
+
+The [contact/retiming package](../p104-joint-neutral-search/README.md) supplies two useful positive controls:
+
+- Its generator discovers the published predecessor helper708 cut after another role's MOVE, without being handed that cut, and emits a legal dyadic positive control on all 753 arbitrary columns.
+- Advancing source MOVE16655 of role153 to just after16649 creates a new legal contact at no extra paid or scalar charge. Exact endpoints, frames, lifetimes and inventory are preserved.
+
+The finite two-meeting searches are retained for provenance, with their complete domains and overlapping coverage explicitly recorded. Their millions of failed trials are supporting evidence, not the headline theorem. The registered publication replay regenerates the contact sets, positive controls, legal retiming and invoice, then checks the six exact three-meeting certificates. It authenticates but does not re-enumerate the historical exhaustive search receipts. The optional predecessor `verify.py --full` remains available for that separate task.
+
+## Reproduction
+
+Python 3.12, standard library, from the repository root:
+
+```sh
+python3.12 -B scripts/research_package.py --package research/p104-boundary-reuse
+```
+
+Both normal and optimized registered commands use temporary copies of the frozen packages, compare every original file after replay (ignoring only documented diagnostic timings and compressed-container metadata), and perform exact cross-package predecessor-byte joins. No separate local worktree or network service is needed for scientific replay. [Publication validation](VALIDATION.md) records the actual assembly checks. Original README and validation statements about local-only status remain historical provenance, not current submission status.
+
+The scientific word is pinned to PR5718 `a1534ac45401c9a9c80c07aa65c89e00e33594f8`. Its archive merge is not admission. At the submission base `de5ec2e7a6074068452f45b1a191b5c72832df4a`, selected conditional κ remains .00310045. Frozen sources, contact lists and results retain their exact bytes and attribution. Duplicate vendored predecessor files are intentional immutable copies, checked against the included full predecessor package.
+
+## Significance for researchers
+
+More contacts and favorable information spans do not necessarily compose into a legal temporal program. The contact generator and retiming are reusable construction tools; the exact proof explains why longer local scalar words at these same meetings cannot earn the proposed `18(H4+H5-H9)` target credit. The complete source bill stays 66,348 paid jobs and 12,976 productive fields.
+
+## Open next questions
+
+Reopen only with a named contact/delivery change that defeats the two-direction contradiction, a different cut basis, or an explicit owned-memory contract. Extract its causal transfer before starting a coefficient census and price its full acquisition, delivery, restoration and cleanup against the complete target allowance. No surviving construction or favorable complete bill for these changes is supplied.
+
+Credit [flyingviper](https://github.com/flyingviper)'s [PR5718](https://github.com/CrocSwap/integer-mult-bounds/pull/5718), [Ali Romman](https://github.com/aliromman)'s [PR5713](https://github.com/CrocSwap/integer-mult-bounds/pull/5713) and [PR5708](https://github.com/CrocSwap/integer-mult-bounds/pull/5708), and the full native, helper, controller, compiler and analytic lineage retained in the original notices. Prepared for [zxv](https://github.com/zxv) with substantial OpenAI Codex assistance. Apache-2.0. No full native/outer admission, formal proof-kernel check or independent human theorem acceptance is claimed.