PhotonHashv0.4.0-shadow.1 / LF2-K8Non-consensus candidate

Proof of work
shaped for light

Exact low-bit matrix arithmetic, mineable on electronic hardware today and structured so photonic acceleration could compete later. Evaluated separately from the operating shadownet.

Specification ↗ Benchmarks
01 · Construction

One vector, eight matrices, one digest

Each attempt derives one low-bit vector from the nonce. That same vector is multiplied against eight independent 4096 × 4096 matrices derived from the parent block. Results are requantized exactly, XOR-folded, and hashed once. The passes run in parallel; no pass feeds the next.

NONCE →x
x · M₀requantize
x · M₁requantize
x · M₂requantize
x · M₃requantize
x · M₄requantize
x · M₅requantize
x · M₆requantize
x · M₇requantize
⊕ XOR FOLDdigest
02 · Specification

Exact by definition

The portable integer implementation defines the result. Every accelerated implementation must reproduce it bit for bit. In Laser Core, the optional CUDA helper's candidates are rechecked on the CPU.

STATUS
Frozen non-production candidate under active qualification
WORK PER ATTEMPT
134,217,728 exact low-precision MACs
MATRIX BANK
Eight parent-derived 4096 × 4096 matrices · 128 MiB unpacked
NONCE INPUT
One nonce-derived vector shared across all eight passes
REFERENCE
Portable exact integer implementation with independent implementation vectors
03 · Electronic benchmarks

How much of the work is matrix math

The measured share of GPU execution time spent in matrix multiplication. A higher share means more of the workload is the operation optics could accelerate. It is not a speedup, an energy saving or an optical result.

v0.2.2 · HISTORICAL BASELINE
11–12%

Reproduced reference outputs, but the profile missed its hardware-economic target. Kept as a baseline.

RECURRENT R1 · ARCHIVED
93–94%

Needed an exact nonlinear boundary between rounds. Modeled conversion energy and noise sensitivity ruled it out.

LF2-K8 · CURRENT · RTX 5090
61.55–61.98%

Measured GEMM-time share. Under independent implementation, adversarial, cross-hardware and network qualification.

04 · Unresolved

What qualification must rule out

Ruling out a catastrophic electronic shortcut is a production prerequisite.

Pass omissionApproximate outputsPrecomputationBatchingFusionMatrix-bank amortizationFPGA implementationsASIC cost modelsCross-hardware agreement
05 · Path to physical hardware

Measured end to end, or not claimed

Photonic advantage is plausible and engineered for, but not demonstrated. Any superiority claim needs complete-system energy and throughput measurements covering:

ModulationDetectionConversionControlCalibrationHashingRecheckInvalid trajectories

Hardware teams: start with the repository ↗ or reach @photonicmoney ↗.