[aliromman]

Three independent helper inputs can be retired through scalar contacts at different times. A retained owned buffer collects an input, exchanges it with a fresh arbitrary input, and injects the difference through later contacts. Four paid query subdivisions provide the required access. This changes the simultaneous-meeting contract of PR5751.

On exact PR5774 at 7d6ec0cdef31d0c37b7d2563e709979caa0d5eae, the complete conditional saving rises from 0.00312192127 to 0.00312265044: +0.00000072917 (0.0233565%). The exact raw crossing is (0.00312265047577, 0.00312265047578). This is a small gain and does not meet the donor's one-percent target. The larger investigation continues.

The package research/p104-asynchronous-helper-hubs contains the temporal proof, fixed rational plan, exact constructor, literal word, source export and complete price. Two rank-nine targets become rank-five/rank-four pairs; a third becomes rank-eight/rank-one. The full caller has 13,444 productive targets, 66,906 paid calls and rank surplus 760. It retains every local call, thin query and whole idle. All buffer initialization, storage, scalar operations, native queries, clearing and routes are charged.

Validation on head 6b936acd68a7e0f9062037900f0fe9bc4b6fdd25:

• Primary and registered normal/optimized replay pass: all 795 input columns, 632,025 coefficients, exact temporal histories, separate dense transpose, 2,926 native maps/phases, 1,124 helper normalizers, 208 main adapters and the full physical caller.
• Both rational price engines, completion, compiler, 54 strict outer predicates and four functional checks pass. Ten word/caller/invoice controls and the native phase control reject.
• Metadata, source pins, committed bytes, official export and rendered README pass. A separate read-only, Git-free replay under -I -O -B with empty PATH passes in 115.057 seconds. Every payload byte and mode stays unchanged. Export SHA256: 9c788f45c96fa6d593a82df13f809ef72016fa5d60109627c037c8641f2bf8a0.

The five inherited framework assumptions remain. The large uniform word and rounded multiplier are not materialized. Full-history preflight was not run in this shallow checkout. Independent acceptance and hosted scientific CI remain pending. The selected frontier is unchanged; no other PR is merged or closed.

Base main: de5ec2e7a6074068452f45b1a191b5c72832df4a, priority revision 8. Fresh complete overlap covered 337 open PRs and 5,161 Discussion records, with sparse-local and unpublished-work limits. Canonical investigation and current scope.

Prepared by Ali Romman with substantial OpenAI Codex assistance; no subagents. PR5774 supplies the retained native copies; PR5751 the earlier paid meetings; flyingviper's PR5718 the distinct boundary bases. NOTICE retains every P104/native/helper/controller/compiler/BIT predecessor, including Joel Pulikkan's dimension-four basis. Apache-2.0.

[aliromman]

Hosted CI update for exact head 6b936acd68a7e0f9062037900f0fe9bc4b6fdd25: the planning job failed with eight failures and two errors in the shared test_research_current_assignment.py suite. The log shows the stale revision-7 expectations (follow_up_steps, catalog_first_step and 16 rewritten cards) against revision8. PR5761 already addresses this shared failure.

The scientific replay job was skipped. No hosted scientific pass is claimed. The package's primary, registered normal/optimized and separate read-only Git-free checks passed as recorded above; this CI result does not extend that evidence. I have left the separate CI repair and selected frontier unchanged.