[Clear1nsight]

The actual selected first F0/B1 pair cannot be replaced by corrected QQ16 plus a post-pair patch confined to its sixteen selected output ports while preserving the other bridge inputs. A source-bound ordinary-port witness gives the next bridge output 0 versus +1/2. This closes that narrow connection experiment; it is not a general QQ16 barrier or a cheaper supplier.

The package binds replica0 calls[0]/calls[4], intervening replicas and borrowed helpers, the pair1 idle transition, and execution6/source-event3's full signed bridge. It replays the actual #4015 decoder word on the 224-dimensional data specialization, distinguishing recovered original values from the caller's current values. It never inverts the private-clearing 712-field transition or identifies F0/B1's different helper banks (intersection40, union920).

For the declared independent-cover module input X0(d,z)=delta[d=I73], constant in payload, ordinary current Y1 is 0 under the selected pair and -1/2 under QQ16. Actual 73D owner products give (D_(rho1 rho76)+D_(rho1 rho81))/4-I/2; identifying cover owners falsely cancels this residual. The excluded adapter preserves ordinary pair0 fields, pair1 bridge inputs and other persistent bridge inputs. Earlier producer changes, repairs writing U, different proved endpoints and separately established reachable-state restrictions remain open. We have not proved this witness is generated by a particular outer multiplication input.

The literal candidate operation ledger identifies sixteen removed H8 centres and thirty-two introduced H1 lines. QQ16 pair versus selected pair adds 28,616 scalar allowance; appending #4015 adds 224 H1, 448 wrappers, 352 maps, 416 CLEARs and 26,944 scalar allowance. The raw subtotal is +240 native calls, +128 native rank sum and +55,560 scalar allowance, before a valid current-endpoint repair. These different units are kept separate. The decoder's 224 owned outputs, conditional terminal cleanup, general repair, physical placement/movement, precision and actual recursive usage remain unpaid or unproved. Complete saving and a new kappa are unestablished.

Only research/qq16-paid-pair/ is added, based on canonical bb4bccbf42844d2801487450727c2f63e10d1d0f. Exact pending heads #4015 0fff81d8997cef646538bcd4f60d7d455cf93b8e and #4044 a8c75dd7c964855cc9b03c210575be27c65467e4 are retained as authenticated data. Their original-head blobs were separately compared byte-for-byte; generic replay uses retained data without fetching sibling heads. Current guidance, priority6, family gate and selected record were checked. The overlap helper reports 37 local records/19 matches but missing remote PR/Discussion coverage; bounded authenticated searches supplement it, not a novelty certificate.

Verification: Python3.12.3 registered isolated replay; exact rational data maps and full73D owner matrices; independently implemented owner/cost replay; real emitted coefficient mutation, route-order and collapsed-route controls; normal/optimized Git-free read-only export; source/vendor/manifest corruption, typed-receipt rejection and overwrite refusal. All270 onboarding tests passed with11 platform skips. REVIEWS.json records the final assisted adversarial review, exact hashes, resolved scope defects and shared context. No independent human review, formal-kernel proof, arbitrary-payload native execution, full selected cold replay or supplier admission is claimed.

Reproduce:

python3.12 -B scripts/research_package.py --package research/qq16-paid-pair

The next bounded experiment is a paid query-factor repair allowed to write U, coordinated with the producer so normalization is fused into work genuinely removed. Bind all eight query channels and ordered owner/payload maps at the actual bridge. Success requires both bridge legs/current state and restoration to match, with a favorable complete bill and justified incidence. Stop at the delta-owner witness, another permitted separating state, or a sound optimistic cost screen that repays the proposed deletion. #4003/#4030's different ordered-query/common-frame restrictions must be checked before reusing those routes.

Canonical investigation: #3985 (comment). Credit #3985/#4015/#4044 and the distinct ownership/numerical/physical audits #4047/#4043/#4042; ywz65/#2707, Arham Tahir/#2681, Ali Romman, Douglas Colkitt and the retained QQ16/native/hosting/controller lineage; Tekalign/tekuuu's distinct ordered-query work. Exact attribution is retained in the package and unchanged source notices. Chris R/Clear1nsight with substantial OpenAI Codex assistance. No merge, selected-frontier change or continuing exclusive claim.

Publication head: f0c9a602d1c82b326917e87397dc7fab8750534b. The official committed-byte/source-history preflight passed for all24 package and declared input files. Scientific review hashes and hosted CI status are separate.

Publication verification completed 2026-10-20 UTC.

Hosted registered workflow 39349478363 completed successfully. Planning job118390011846 passed270 tests(11platformskips) and selected only this package. Replay job118390715884 passed the registered exact replay, normal/optimized Git-free export and rejection controls, and committed-input preservation. Decoded checkout logs bind synthetic merge ad31fe1b0e382ef7b63a0984e8a94dbc40ab8c04 to scientific/publication head f0c9a602d1c82b326917e87397dc7fab8750534b and base bb4bccbf42844d2801487450727c2f63e10d1d0f, CPython3.12.15. Hosted success is a reproduction/integrity result, not supplier or mathematical acceptance. The24-file committed exporter/checker and a fresh read-only optimized Git-free extraction passed; payload hashes stayed unchanged. Export archive SHA256 310d5e779c0456a98488e5fc965c877ba557a0dd6329fe72c25a534178114592. No scientific bytes changed after publication.