Share closed primitive engines to remove parent parking at child calls

[Clear1nsight]

Result and scope

After #4166, avoiding parent parking could appear to pay for QQ16 recovery, but the missing fair baseline was original arithmetic using a compatible external-call protocol. This package implements that baseline with one original numerical pipeline and a second controller bank. The controllers lend the same closed scalar/leaf/monomial/Ordinary objects sequentially, preserving literal allocator history and encoded slot identities. It adds 1,023 fixed controller tapes, with a conservative 66,496 read/write/unit-move action setup/terminal-retirement bill; no second Ordinary engine is constructed.

This is one class-C enabling mechanism plus a quantified complete-root source-action comparison. It is not another QQ16 caller installation, proof that recovery is affordable, unrestricted Turing-step/runtime dominance, all-node recursive substitution, or a new κ. The source/native/ordinary/Compact/hosting/long-record and row-stock premises remain explicit.

Constructive comparison

A retains original selected arithmetic and embedded calls. B retains exactly that arithmetic and uses the shared-primitive external controller. C/D are the frozen #4166 fixed/adopted publication variants. The same external RootLayer ports and continuation boundary are preserved; recursive descendants remain original C.

The exact complex park/restore cost is 56S+22r+11 actions, including loop tests, append checks, markers, empty cases and rewinds. A source-derived injection proves that the full 15-tape capsule has length C<=S, the parked parent body sum. Literal root flags, ancestor stack translation, headers, first-use markers and unmatched terminal returns are handled explicitly; equal ranks alone permit no cancellation.

Across all 510662Γ original root calls, including 493097Γ later calls, the conservative unit-action saving is

A - B >=510662 Γ (52 B_bank+18214)-66496 >0.

B_bank is the complete parent-bank bit length, already including Γ. Separate coordinate bounds are positive too, so every fixed positive weighting of reads/writes/unit moves improves within this source tariff. Separately compiled finite-control costs retain their stated threshold; elapsed timing is not this theorem.

A new guarded internal child-return ownership exchange saves 11n+4 actions per returned n-bit packet, less finite rebinding control. It updates all packet/private/parked/parking/backend-output bindings and preserves public ports. The same optimisation is available to compatible variants. It is distinct from #4166’s already credited whole-bank 11B_bank+4-c_R saving.

What changes for QQ16

The fair decomposition is T_D-T_A=(T_B-T_A)+(T_D-T_B). Avoided full-parent serialization is a generic protocol benefit and cannot be credited exclusively to recovery.

The independent literal census binds every source occurrence, target/rank and installed position. Of 510646Γ retained calls, 503674Γ have matched payloads under #4166’s semantic refinement and 6972Γ are changed-core calls. The matching-payload header component adds 14,102,872Γ actions, including 13,806,716Γ later suffix actions.

A real gap in the proposed separate-engine proof was found: closed native arenas retain free-slot order, with slot IDs encoded in words. Sharing the actual primitive bundle closes this for B. Frozen D’s signed nonheader context residual remains explicit, together with changed-core differences, 336Γ service calls, 32Γ added producer calls, 16Γ removed centres and complete nonchild work. No negative-cost theorem for independent engines is inferred. A paid canonical six-arena reset on new fair variants is the next discriminating experiment.

Evidence and limits

• Authenticated source-before-execution closure, independent full selected literal census, exact primitive derivations and 60 actual codec/tape calibrations.
• Complete nonconstant direct runs over two generations; actual nonleaf child with ten real grandchildren, returned live caller state and next source ADD; both bridge legs and later-call diagnostic; multiple prefixes, spectators and coefficients.
• Primitive-state fingerprints include allocator queues, leases, heads and numerical-codec state. Saved PC/target/root, active engine/parent reentry, stale bindings, alias/head, omitted suffix/constructor/double-counting and premature retirement controls reject.
• A real same-output zero-child losing case is retained: the adapter adds an entry read and positive retirement. The whole-root theorem is restricted to source-admitted nonleaf selected roots.
• Registered ordinary/optimized read-only Git-free export replay, typed/duplicate receipt and archive/member corruption controls; committed-byte/source-history preflight and official committed export checks are recorded in the package.
• Frozen read-only assisted adversarial review reports no unresolved blocking defect in the stated scope. It shares source/checker dependencies and is not independent human or formal-kernel acceptance. Hosted CI is separate.

Production O73 payloads and complete RootLayer.run were not executed. All-size applicability is a written source refinement, not inferred from reduced fixtures. Historical backing survives erasure and role exchange. Selected κ stays 836330098538459/500000000000000000.

Sources and coordination

Scientific main e40a7b454da6a0df53e358aa3265b9c1c39de2c8, priority6; frozen #4166 0b082d22a0e9a442d264ab0633e35ef26c790630 and #4129 e708f02bf78c3ed6d770cedf00724ebc5ce928d1. New review/publication heads are recorded separately. The canonical #3985 handoff preserves prior results and the active checkpoint. Bounded all-state searches and named heads/reviews/Discussions were refreshed; capped/cache-limited searches and unavailable helper remote coverage do not prove novelty.

Credit Jose Manuel/#3385 and retained numerical/physical/Compact/ordinary authors; Sagar Thakrar/#3464 for compatible parent restoration; Tekalign/#3380 for the ownership-exchange idea, adapted at a new site with a new proof/bill; Ali Romman/QQ16, Arham Tahir, ywz65 and #3985/#4015/#4106/#4109/#4129/#4166. All requested compatibility decisions and original notices are retained. Chris R/Clear1nsight with substantial coordinated OpenAI Codex assistance. At most this one integrated PR; no frontier or merge action.

Final local publication receipts at 6749e458800f461535d64cc91c017552d9f55e5f: registered primary and all 6 boundary groups PASS (252.471seconds); committed-byte/source-history PASS for 106 files; official export/check PASS for 106 payload files plus manifest. Actual committed read-only Git-free ordinary/optimized replays PASS in 124.643/122.214 seconds, with empty stderr and unchanged contents, modes, mtimes and inodes. Archive SHA256 b9ffc464b72905fa1f789782ebab06269e1e0bff9d7342eed19903f4d51a4518. Metadata clarification was checked after the scientific freeze; executable/proof bytes are unchanged. Hosted CI is pending at publication and is a separate status.

Start with PROOF, whole-root comparison, resources and canonical handoff/history.

Assigned task joint-query-controller, scope key 42581d429acc9509d503; scope changed to the specifically assigned physical-protocol continuation after #4166. The original card remains pinned in TASK.json. This mechanism resolves the actual caller/context and budget bottleneck; it is not an unrelated task selection.

Submission refresh: canonical main 12fa8c5c844be21bfefb38a19503fbd867a11244, live priority revision 7, was fetched and read before submission. Its updated QQ16 card explicitly asks for the #4166 whole-caller protocol comparison, retaining every later root call. Standing instructions, selected result and physical controller are unchanged. The scientific priority-6 base/card and source inputs remain frozen; this class-C result stops recovery expansion at the explicit unpriced threshold.

Open next questions

The smallest next experiment is fitted Ordinary execution from two different legal closed arena free-slot orders, followed by a paid canonical six-arena reset or a literal codec-equivariance proof. Include every validation and finite-control charge and preserve actual borrowed bytes/heads. That can close the same-payload native-context residual for explicitly named new variants. Then price the 6972Γ changed-core calls, complete recovery service and nonchild difference against fair B; only a favorable complete comparison justifies a recovery-specific benefit.