[ywz65]

PR2511 supplies a complete logical selected-sum decoder, while PR2512 supplies current-frame L2 reads. This composes them into the complete added local native decoder, including early-record capture, temporal ownership, all remaining reconstruction, source restoration and record retirement. It returns all 224 original logical arrays in the common physical frame F_U1. That output frame is explicit; it does not claim restoration of the original pre-F0 physical frames.

The exact added invoice is 224 rank-one + 16 rank-eight children (native mass 352), 480 native wrappers, 208 full-cover maps and 17,472 real scalar primitives. The scalar bill comprises 856 unit ADDs, 832 quarter ADDs and 320 CLEARs, including 160 primitives for early capture and 32 for retirement. The inherited F0/B1 calls keep their own costs; the word's history marker is not a free gate.

Sixteen separate owned early banks capture selected X after F0 and remain unchanged through B1. Their values are copied into the eventual selected X outputs, which are upgraded once to F_U1. Eight of those owned outputs feed the L2 correction and remain available. The actual F0 owner chart is identity, so a complete local inverse recovers all original X with sixteen local scratch arrays. Original selected Y is built directly in the eventual Y outputs from the paid current sums; no separate selected-sum bank is allocated. All 224 borrowed current physical X/Y arrays restore, all sixteen early banks retire, and all private scratch returns to zero. The caller supplies 224 owned full-cover output banks, 16 additional early banks and at most 16 local scratch arrays per owner. No storage saving is inferred.

The package is research/qq16-complete-native-decoder/. Its proof and literal word track 240 independent temporal source symbols (16 earlier and 224 current) and arbitrary owned-buffer dirt, checking all 224 output operators against authenticated PR2511 logic. The retained logical decoder also checks both inverse orders on its actual 224-input and compressed-observation spaces. That machinery is explicitly reused with credit, not presented as an independent implementation.

Native evidence covers 104 actual translated maps, 281,008 complete phase-character comparisons, 114,688 signed line-basis columns and the retained 65,536-coefficient rank-eight kernel check. Twelve semantic controls reject capture after B1, missing upgrades/normalization/restoration, wrong owner payload shifts, an inverse sign, output dirt, borrowed clearing, leaked scratch and unretired early records. Six boundary test groups cover early retirement, private-sweep chronology, complete word/receipt requirements, source authentication before execution, Git-free read-only export and overwrite refusal.

Validation on Windows, standard library only:

python3.12 scripts/research_package.py --package research/qq16-complete-native-decoder
python3.14 -I -B -O research/qq16-complete-native-decoder/verify.py

Native CPython 3.12.7 registered primary and six test groups pass, and 3.14.5 optimized primary passes. A separate AI reviewer found no material issue, passed the optimized3.12.7primary, and independently reimplemented the JSON-only scalar/ordered-owner replay: all224outputs with5840nonzero terms over240temporal symbols,104source-registry polarization maps, borrowed logical restoration, early capture/retirement and the complete invoice agree. This independent checker does not separately lower the native kernels or recertify the original full native F0/B1 words. No human review, proof assistant, hosted CI or selected supplier cold replay is claimed.

Physical retention, whole-cover movement, child hosting, scalar addresses, precision, use of the 224 outputs, global controller placement and all-size admission remain open. This is a complete finite added decoder invoice, not a favorable physical/supplier budget, minimum native count, new retention theorem or new kappa.

Base is freshly fetched canonical main 0fdd7d82de03e43c9e5a1000dfbcd380d1b687d2, priority revision 4. Source PR2511 is pinned at f40e74a27f7c8781feea06a1571dc99ed5d9a004; PR2512 at 851dd3486a2d0549b51e85978a65ccb4eec619c0. Both heads and main were verified unchanged at 13:46 UTC. Exact predecessor bytes, Git blobs and SHA-256 pins are retained. The task-picker seed 6c82be10ab151cf0e67b46fabec4b461 selected recursive interfaces; the card scope is narrowed to this concrete missing decoder contract. Baseline bindings passed.

Overlap coverage is bounded: complete cached 679 open PRs at 12:25 UTC and 1,197 Discussion records at 12:29 UTC, supplemented with the latest 50 updated open PRs via REST at 13:30 UTC and targeted source/claim checks. GraphQL quota prevented a newer complete Discussion refresh. Advance claim.

Credit ywz65's PR2511/2512 composition sources and PR2456/2468/2492 machinery, thakrarsagar/PR2470 for L2, starship-s/PR2443 for the original cut/F0 inverse, Ross Richardson/PR2480 for fixed-L2 retention, Ali Romman for QQ16 and the retained native/geometry/supplier lineage. Source notices remain. New work by ywz65, with substantial OpenAI Codex assistance in derivation, source inspection, implementation, testing, writing and review. Apache-2.0.