Pair QQ16 recovery sources to remove 100 paid unit additions

[Clear1nsight]

The full QQ16 future-query circuit can share paid sums between sources with the same coefficient support. This changes the arithmetic DAG and removes 100 unit ADDs / 800 scalar allowance against #4015, after charging actual source acquisition, initialization, response and output restoration.

The complete 3,713-event word returns all 224 original outputs in F17 and preserves all 224 borrowed current X/Y values and frames, with arbitrary initial owned dirt. All 104 corrected W sources remain, including every T contribution and inverse-prefix effect. The 52 pairs use 316 query ADDs instead of 416: 48 pairs save two each and four save one each.

Complete local quantity	#3985	#4015	This word
Unit ADD	1920	1920	1820
Quarter ADD	896	896	896
CLEAR	536	416	416
Scalar allowance, inherited 8/12/2 tariff	27184	26944	26144
H1 / wrappers / cover maps	224 /448 /352	224 /448 /352	224 /448 /352
Additional owned banks	112	0	0
Whole-word private peak	16	16	16

Eight query roles and two separately normalized sources occupy ten existing deferred Z_U output slots. All ten paid CLEAR/X-copy definitions are restored before the exact original suffix. The tenth host does not create an extra CLEAR credit: saving 800, not 802. All 224 owned outputs remain required. #4007's separate materialized schedule is compared as published evidence, explicitly not re-executed or physically dominated.

Evidence and reproduction. PROOF.md, separate construction/acquisition proofs, WORD.json and complete typed RESULT.json give the paid boundary. Runtime inputs are 22 SHA/Git-blob authenticated canonical files, bound before inherited execution. Pending comparisons retain only WORD data; credited replay/native routines are copied locally and authenticated, without network or sibling-checkout imports. Five reused predecessor files were independently compared with their exact original-head Git blobs. PREDECESSORS.json and REVIEWS.json retain source heads, attribution and assisted review scope.

Local CPython3.12.3 registered isolated replay PASS, outcome partial: complete original/checkpoint and arbitrary-current routed recovery, all phase frames and borrowed return; actual residual Q2(X_T)−M21 Q1(Y), nonconstant witness, attainability and response visibility; 114,688 signed-native basis columns with 224 occurrences bound; separate two-symbol identity and full acquisition/event recount. Controls reject premature destructive addition with an unchanged invoice, missing T/source initialization/response mapping/tenth-host restoration and private/native overlap. Normal/optimized Git-free read-only export passed in 158.494 s; source/copied-executable/manifest/vendor/whole-word/type/duplicate-key/overwrite boundary groups passed in 168.131 s.

Publication head 4852991. Official committed-byte/source-history preflight and export/check PASS for all 46 package/input files; whitespace check PASS. Hosted registered verification passed in run39350712039, replay job118394540764. Completed logs bind synthetic merge c980b84 to this exact head and canonical base, using CPython3.12.15. The primary verifier passed; normal/optimized read-only export passed in114.825s; the three remaining boundary groups passed in116.164s; the registered wrapper reported PASS outcome=partial; committed inputs remained unchanged. Planning passed270 helper tests with11 platform skips and selected only this package.

The official committed export also passed a separate complete optimized read-only replay with no Git directory and PATH=/nonexistent; all47 bundle files (46 package/input files plus the export manifest) stayed unchanged. Archive SHA256:13b1320eb0d18c98f3020dc004791bce904ec9b77c9e0a89ca03c3bcb3a9208d. These are reproduction and integrity results, not supplier or mathematical acceptance. Scientific package bytes have not changed after publication.

python3.12 -B scripts/research_package.py --package research/qq16-sequential-recovery

Scope. This is an exact local scalar-allowance improvement, not measured runtime or a completed supplier. Reassociation changes ranges and rounding; #4043's identical-trace transfer does not apply. Physical movement/control, tag ABI/native exclusivity, numerical range/error, actual invocation/coverage/idle credits, weighted recurrence and both supplier caps remain unproved. No κ improvement. #4105's selected-only actual-bridge obstruction remains a distinct caller limit; this recovery circuit does not supply its missing connection. Tekalign/tekuuu's #4030 common-frame gate is respected: all 104 source normalizations remain.

Task and overlap. Changed scope of joint-query-controller under the explicit recovery-arithmetic assignment, scope key 42581d429acc9509d503. Canonical base bb4bccb, priority 6, current guidance/family gate refreshed at entry/pivot/submission. One bounded pivot followed a failed identical-projection screen. Targeted all-state QQ16/sequential PR and Discussion checks supplement a local-only overlap helper; broader remote coverage remains partial. Canonical investigation. Only this new package changes; frozen inputs and selected frontier are untouched.

Next experiment. Derive exact intermediate norms and output-error propagation on this frozen paired WORD. Under a stated numerical/bit-cost model, test whether extra precision preserves the 800 scalar-allowance advantage; stop with the precise repaid budget if it fails. Physical and actual caller connection remain separate prerequisites for κ.

Credit. #3985 owns the preceding 336→224 native reduction; #4015 owns streaming/nine-output hosting; #4003 supplies the ordered residual. Preserve ywz65/#2707, Arham Tahir/#2681, Ali Romman, Douglas Colkitt, commdata2338/#2651, Nick Petalas and retained native/compiler/geometry/supplier lineage. Tekalign/tekuuu's #4030 normalization gate is a separate contribution. Chris R/Clear1nsight with substantial OpenAI Codex orchestration, construction, separately implemented accounting and read-only adversarial assistance. Full replays share canonical algebra and #4015's engine; independent execution/inspected derivation is distinguished from an independent full decoder implementation, human review and formal-kernel acceptance. No merge or frontier change.