[aliromman]

A read from owned centre 747 is replaced by four scalar updates at two other contacts. Field 715 is not read between those contacts. Their exact source-functional sum restores it before the next read, so its H1 and H2 transports merge into one H3. Every independent target and centre query remains paid.

The complete conditional saving is κ = 0.00310915222, up 0.00000043910 (0.01412482%) over PR5684 at d0f6fa2555410a1e23fe4191424c2097ab3d5d97. The whole invoice changes by -18 H1 -18 H2 +18 H3, with no target change: 66,348 paid calls, 12,958 targets. The raw crossing is strictly in (0.00310915224562, 0.00310915224563). This is a new distributed-update construction with a small gain; it does not deliver the requested large breakthrough.

The package contains the exact temporal proof, full chronological word, source pins, native, caller and complete price receipts, and open next questions. Publication head: a5348f7143f95fe0046b92b96172e30763c7d731. It changes only research/p104-temporal-updates/. It does not combine pending PR decimals or change the selected frontier.

python3.12 -B scripts/research_package.py --package research/p104-temporal-updates

Validation covers all 565,504 public/owned endpoint coefficients in rational and dense transpose replays, 2,917 native maps including absolute phase, the actual physical caller, two exact moment engines, completion/rounding/tail guards, the finite compiler threshold and all 54 strict plus 4 functional outer checks. Nine semantic controls reject altered source signs, a missing update, a read across the claimed interval, false frames, an unpaid query, an extra target and omitted native/owned operations. A damaged frozen source rejects before import. Normal and optimized registered replay passed. All 21 committed package files passed the byte checks and official export checks. A read-only optimized Git-free replay passed with an empty PATH; all 22 export files, including the manifest, kept their exact bytes and modes. The sealed scientific commit c1ec999229213f052e899d89e2e3dd696ceb33ef is an ancestor of the publication head; only PROOF.md and VALIDATION.md changed afterward. Validation details separate author checks from hosted CI and independent review.

The five inherited native, owned-erasure, compiler, precision/tape and analytic/recovery assumptions remain. The huge uniform word and complete rounded multiplier are not materialized. All eight H7 centre queries remain; this deletes one recipient contact, not a centre query. No independent human or proof-kernel acceptance is claimed. Full-history ancestry preflight is unavailable in the shallow checkout.

Coordination: canonical investigation. Main was refreshed to df0137efd35e11b4445734bf612bdd4c7ebc330c, priority revision 7. The 08:27 UTC overlap refresh covered all 392 open PR bodies and 5043 Discussion records. Sparse local coverage was 9 of 7676 notes; arbitrary diffs and unpublished work remain outside that coverage. The scoped matches cover older mutable-source work and separate G20/helper-retirement tasks. No matching active P104 distributed-update claim was found. The final submission checkpoint rechecked canonical main and the live PR/Discussion delta, including the distinct G20 work in Discussions 5710/5711 and the released P104 helper-cut scope in 5680. PR5708 is a separate ordinary BIT component result; it is not used or added to this whole kappa claim.

Credit Joel Pulikkan/GamingPuzzled with Claude assistance for PR5684; Ali Romman's PR5678/5660, complete caller and interface lineage; zxv, Cao Yifan, Cinna Davis, Jacob Alan Sussman and every retained native/helper/compiler source notice. New construction, proof and verification: Ali Romman with substantial OpenAI Codex assistance; no subagents. Apache-2.0. No global literature-priority claim.