Mouillay Quantus GPU Miner TEST 0.1.25-test26
Based on Quantus Network quantus-miner (Apache License 2.0).
Upstream commit: 8e2d27219c93567fca67bf049f52882f6c87df35
Copyright and license notices in LICENSE and THIRD-PARTY-LICENSES.txt apply.
Mouillay changes: exact canonical Goldilocks GPU arithmetic, embedded native SM86/89/120 kernels and new TEST CLI / independent Quantus TCP pool transport.
Protocol facts checked against 1640675651/CPPminer MIT implementation commit ce8712a06fc38d73d6961070558ac2d1874e2728; no code incorporated from restricted KanQ license.
Current TEST26: developer fee is 0.5% on poolmouillay.com and 1% on other pools, based on completed active scan time. Developer identity: qznRtYqTb5DUwCNgg7CzyDpAryM16wMGSdQxXDfMEwBeA1bHS.devfee. Pool fees are separate.
Current SM86 default includes original unmodified Peak GPU modules; see PEAK-KERNEL-PROVENANCE.txt.

Historical development notes below describe earlier TEST versions, not current fee/backend settings.

TEST3: independent affine preparation tables and full-nonce reconstruction with exact CPU verification; commercial extracted kernels are excluded.

TEST4 adds independent exact net-carry product reduction and a serial ordinary x7 chain. No developer fee.

TEST5 adds exact specialized96-bit reduction and shared-memory spill storage on SM86. No developer fee.

TEST6 adds an independently proved conservative final-product-limb prefilter. All survivors retain exact complete verification. No developer fee.

TEST7 retains exact conservative filtering with sparse/zero external constants and cold local spill storage. No developer fee.

TEST8 adds an independently authored exact SM86 SASS prefix body, with no competing miner kernel included. No developer fee.

TEST9 adds exact fused diagonal products, paired Sbox scheduling and one-batch pinned result transfer. No competing miner kernel is included. No developer fee.

TEST10 adds an independently proved conservative final-product interval gate, with exact completion of survivors. No competing miner kernel included. No developer fee.

TEST11 adds independently proved exact carry-flag96-bit folding and removal of overwritten MDS copies on SM86. No competing miner kernel included. No developer fee.

TEST12 retains independently authored dependency-aware SM86 scheduling and exact lowword-only final-gate products. Experimental mixed Montgomery arithmetic is not selected. No competing kernel included. No developer fee.

TEST13 adds independently derived exact affine constant transformations and native multiply-add carry fusion on SM86.104native instructions removed. No competing kernel incorporated. No developer fee.

TEST14 adds an original exact carry-fused Goldilocks reduction and dependency-checked SM86 retiming.1436arithmetic additions removed. No competing program or kernel incorporated. No developer fee.

TEST15 adds an original exact public-bias constant accumulator and SM86 GPR WAR scheduling.88 literal MOV instructions removed. No competing machine code incorporated. No developer fee.

TEST17 follows experiments16-20 and retains the exact direct shared-MDS operand layout:132 register copies removed. No competitor program incorporated. SM89/SM120 unchanged.

TEST18 adds an original public-bound packing of the exact96-bit common MDS addend into two64-bit partial-product accumulators.88 further zero MOVs removed; no approximation. No competing kernel incorporated.

TEST19 qualifies a one-cycle local scheduling reduction at528 SM86 diagonal correction sites; arithmetic and six-cycle carry waits preserved. No global latency assumption or competing kernel incorporated.
