Construct a delayed-query P104 side word with full dirty restoration

[jmbarrancoml]

I construct a finite forward P104 identity word by feeding four delayed queries into the retained selected side producer and taking four more queries from omitted target rows. The word preserves COPY56 and restores all 628 arbitrary dirty helper inputs. It replaces the separate gather/delivery architecture, so the frozen-itinerary and terminal-correction obstructions do not apply unchanged.

The registered Python 3.12 replay passes all 836 persistent input columns and eight private columns, checks the retimed common frames, and verifies 241 new native recipes covering 421 paid occurrences, including global phases. Three semantic controls and a native phase control reject faults. The complete local profile has rank mass 6782: persistent progress 6726 plus COPY56.

The fixed profile does not improve the selected saving. I prove this for every paired-chain stage count and any number of fully paid auxiliary families under the stated coverage/halving model. This is a reusable local forward construction with a priced limit; global chronological hosting, a dual orientation, precision/tape bounds and all-size admission remain open.

I started from canonical main 0fdd7d82de03e43c9e5a1000dfbcd380d1b687d2, read priority revision 4, and coordinated in Discussion2121. The final live refresh covered 391 open PRs and 802 Discussion records with complete pagination. I read the neighboring 2131/2123 scopes and kept the changed producer distinct.

I retain credit for Ali Romman's source reconstruction, the selected Hamming/native lineage, Rohan Arun's and Chafik Boukhalfa's P104 work, Adam Gall's and Josie Elliston's delivery geometry, and Douglas Colkitt's interval arithmetic. SOURCES.json records the exact pins and nearby obstruction authors. No selected or frozen source changes.

Author: Jose Manuel (@jmbarrancoml). Substantial OpenAI Codex assistance in construction, proof, implementation and prose; no independent human review.

[Tairitsu-Aya]

CLAIM — bounded independent replay and boundary/budget audit of this PR; expires 2026-10-14 22:00 UTC.

I will review exact head 1df674ca8e9d9e0fb98ba4df402b648e9b3be0a4, focusing on the new delayed-input/late-delivery word, arbitrary helper restoration, common-frame/native-phase coverage, and the fixed-profile all-stage cost screen. I will keep the distinction between a finite forward construction and global supplier admission explicit. No changes to your branch or frozen inputs are planned.

Method: reproduce the registered Python3.12 entry point in an isolated checkout, inspect the formal-row and native recipe coverage, and use small independently derived algebraic/budget controls where they test a concrete boundary. At most three local CPU workers, a 1 GiB replay-memory stop and 20-minute first-wave compute limit; a timeout is a limit, not a defect. The remaining review window is for checking the written reductions and reporting any specific failing example, or a scoped no-finding result. This does not claim to discharge hosting, dual orientation, precision/tape or all-size obligations.

Canonical main remains 0fdd7d82de03e43c9e5a1000dfbcd380d1b687d2, priority4. Your current head has no review/comments and the targeted live delayed-query search returned this PR; the broader complete overlap snapshot at21:06UTC covered403 open PRs and826 Discussion records. I am the contributor to the retained-interval separator in PR2136; your changed source response/route construction lies outside that theorem, and this review will not apply its bound as though your original transport obligations were retained. Review for Tairitsu-Aya with substantial OpenAI Codex assistance; no independent human review is implied.