Consolidate six circuit PRs: 31.72% fewer bit ADDs and 22–25% fewer complex primitives

[alluri-ritual]

The pending circuit PRs contain substantial savings but use different scalar words, temporary bases and precision assumptions. This batch retains six complete research packages and supplies one source-pinned, executable integration with actual-word native and whole-owner replay. It also composes the raw factory changes in this PR with the cheaper complex lowering.

Status: finite circuit integration candidate; no new κ.

Measured operation bill	Before	Consolidated	Reduction
R420 local XOR additions	26,916	18,378	31.72%
Complete hosted bit instructions	8,603,712	6,144,768	28.58%
Selected CK forward real-component primitives	224,938	169,618	24.59%
Selected JP8 forward real-component primitives	257,232	200,380	22.10%
Raw complex factory forward primitives, composed	170,488	123,768	27.40%

The hosted bit saving is 2,458,944 instructions across 288 local invocations. Inverse-transpose complex counts are also checked: CK 224,986 → 169,634; JP8 257,280 → 200,396; raw factory 170,536 → 123,784. These are different bills, not an additive practical-runtime claim. The selected conditional κ remains 0.001635527380375392.

What is consolidated

• Research: compose R420 circuits with 16,704 fewer hosted instructions (no new kappa) #688, Ali Romman: integration/replay infrastructure and yongjae321's Research: simplify one R420 block from 50 to 33 XOR gates (no new kappa) #670 / x0hu's Research: reduce R420 block 5 from 46 to 34 XOR gates (no new kappa) #674 phase-B circuits.
• Research: shorten R420 phase-A quad 13 from 19 to 12 XORs (no new kappa) #691, mtpsl: complete phase-A quad reduction.
• Research: save 6,934 R420 XORs with a restored helper basis #694, Greer Guthrie: all 201 helper-basis pairs.
• Research: save 3,150 R420 XORs with a restored target basis #713, Scott Woolsey Biggart: target-row factoring, recomputed on the helper-transformed matrix; 56 pairs in this composition.
• Research: save 114 R420 XORs by sharing restored decoder operands (no new kappa) #708, Sagar Thakrar: 23 of 27 decoder-sharing pairs in the combined word; the complete original package is also retained and independently replayed.
• Research: lower selected complex scalar primitives by 22–25% (no new kappa) #696, Sagar Thakrar: full selected CK/JP8 primitive lowering, additionally checked with this PR's fourteen raw factory replacements.
• Consolidate six circuit PRs: 31.72% fewer bit ADDs and 22–25% fewer complex primitives #681, Alluri: the complete raw factory circuit contribution remains included.

The six imported research packages are preserved byte for byte at their original paths, with original notices and AI disclosures. Their two separate workflows are represented by one combined activation-ready template. Exact heads, changed-file coverage, attributed mechanisms and proposed post-merge dispositions are in research/supplier-circuit-consolidation/DISPOSITIONS.json. #670/#674 are retained circuit subsets, not claims of complete-PR coverage. Earlier #686 is credited for its compensation mechanism and semantic D contract.

Integration work beyond collecting the PRs

The savings are recomputed, not added from PR headlines. Target factoring is applied to the new matrix after helper factoring. Using all 27 decoder pairs gives 18,370 ADDs but needs 105 bits, so it is rejected under the existing < 2^104 cap. All 27 leave-one-out variants also fail. The recorded 372-extension compatibility pass retains 23 pairs and passes at 104 bits.

The original literal H2 checker is not weakened. The new semantic D-boundary proof preserves every arbitrary target/helper input at frame zero. Original H1–H5 and their conclusions pass on the canonical D with the changed A/B/C, and unchanged native/hosted checkers consume the actual new word.

Precision tradeoff is charged: the global raw prefix rises 189 → 196 bits, so its allocation rises 192 → 256 bits. The two-level bound rises 416 → 432 bits within the unchanged 512-bit allocation. All 120,960 whole helper invocations, zero-bank allocation, recursive profiles and the bit exponent remain unchanged.

Reproduce

# Python 3.11+, standard library:
python3 -B research/supplier-circuit-consolidation/verify.py

# Complete C++17 / Boost / NumPy / SciPy / SymPy hosted replay:
python3 -B research/supplier-circuit-consolidation/verify.py --hosted --out /path/to/new-directory

Expected full result:

PASS_SUPPLIER_CIRCUIT_CONSOLIDATION bit_ADDs=26916->18378 saved=8538 hosted=True kappa=unchanged

The output includes the actual replacement bit word, native and hosted receipts/invoices, and explicit selected/raw complex primitive streams. The opt-in primitive artifacts preserve non-scalar events; their counts and sequence hashes match independently replayed streams. Normal commands compare every deterministic receipt field.

Validation on macOS arm64, Python 3.14.3 and 3.11.15, Clang 16 and Boost 1.85:

• All 660 bit and 716 complex arbitrary-input columns; complete source/helper restoration, exact frames, paid profiles and precision.
• Fresh unchanged C++ legality and 5/9/11-stage replay: 900/1,380/1,620 columns and 96 bridge-omission controls across both words.
• Complete parent/hosted/invoice replay: 17 parent + 16 hosted controls per word, both exact moment engines, 2,458,944 fewer emitted instructions.
• Original component checks, nine integration controls, isolated Python 3.11, minimal-source relocation and five package corruption/output-preservation controls.
• Full selected baseline cold replay (verify_frontier626.py) with no reused outputs, independent owner/interface checks and exact assembly binding.
• Selected-record, reconciliation, imported-source coverage and whitespace checks. EXPECTED.json, HOSTED.json and retained validation evidence record the results.

The included CI template provides Linux Python 3.11/3.14 local checks and a full Linux hosted job. CI activation requires workflow-authorized access: GitHub rejected the connected OAuth token's active-workflow push because it lacks workflow scope. The template is retained at research/supplier-circuit-consolidation/ci/supplier-circuit-consolidation.yml for copying into .github/workflows/. No remote CI pass is claimed. Full repository-wide make verify and Lean were not run; no selected generator is edited.

Review and resolution boundary

Exact main base: 66849e1475c930fd7732b1a0067ffee7e1e95f60. Integration commit: e536f6fc8aad65adef8ac4682d13625598a5c3f9. The source manifest pins 235 consumed files. A fresh minimal archive of this commit is used for final replay.

This is one integration PR. After merge and a fresh head/coverage check, maintainers can resolve the six retained research PRs with their credit and scope linked. Historical ledger dispositions remain unchanged until publication; this batch performs no merge or closure. A final metadata refresh through #730 found the six imported heads unchanged. Later #716/#722/#723 are credited and left outside the frozen batch because their different matrix or private-work/prefix boundaries require another composed recount.

The selected frontier and upstream/ are unchanged. Raw factory edits are not transported through optimized selected provider recipes; selected complex rows use only the checked lowering. The hosted price's PR484 outer calculation is a historical control, not a new selected assembly. Inherited all-size compiler/common-label/mutable-basis, physical hosting, precision/tape and analytic assumptions remain. This is not a full multiplication theorem or runtime benchmark.

Prepared for Alluri (@alluri-ritual) with substantial OpenAI Codex assistance in integration, implementation, proof analysis, verification and writing. Apache-2.0. No independent human or delegated-agent review is claimed. The original circuit, supplier, compiler, emitter, audit and hosting contributors retain attribution in all imported notices and the integration README.