Implement a native local read of the QQ16 future query

[ywz65]

Construct the missing local B2 C read requested by #2443 at its post-B_1_2 completed-X cut. The word restores all 96 source arrays and returns eight query coordinates in explicitly owned output arrays. Future-Y alignment on these X fields leaves a nonidentity native line factor; one inverse line before the read and one positive line after it give an exact local repair.

The complete local invoice is 192 rank-one children, 384 native monomial wrappers, 192 full-cover maps, 384 signed ADDs, eight CLEARs, and eight additional live output arrays. The scalar primitive allowance is 3,088, excluding child/map costs. Owned output CLEARs intentionally discard prior contents; the output arrays are not borrowed helpers. This is neither a universal lower bound nor the full two-cut decoder or a favorable global supplier budget.

research/qq16-future-x-query includes a proof, the literal 776-event word, exact phase/content replay on 104 independent inputs, and all signed rank-one recipes. Eight semantic controls detect omitted normalization/restoration, translation faults, dirty output, coefficient corruption, an early read and a wrong inverse line. Global child hosting, retained observations, physical address costs, recursive precision and complete controller billing remain open.

Validation passed on Python 3.12.7: python3.12 -I -B scripts/research_package.py --package research/qq16-future-x-query (normal/optimized replay, 98,304 exact native kernel columns, five tests). Direct optimized verifier/tests also pass on Python 3.14.5. python3.12 -B scripts/research_baseline.py passes source bindings only. git diff --check is clean.

Direct base: 0fdd7d82de03e43c9e5a1000dfbcd380d1b687d2, priority revision 4. SOURCES.json pins retained inputs. PREDECESSOR.json and the preserved README pin #2443 head 545c05e17243c962b112b91e7f84b4c1a41de7b8. Credit: starship-s / #2443 for the exact access handoff; Ali Romman / #1969 and the retained QQ16 lineage for native conventions. The established 88-read decoder optimum is unchanged. Prepared by ywz65 with substantial OpenAI Codex assistance; no independent human or proof-assistant review claimed.