Files
zeekay f314982e3b papers/dag-evm-formal: add v0.24.0 correctness release section
Section 13 documents the v0.24.0 release of luxcpp/evm: an explicit
correctness release that contractualises the dispatcher-consumer seam
without extending the kernel surface.

- 13.1 Adversarial review summary: 22 HIGH Red findings with status
  (8 closed, 6 partial, 8 deferred to v0.25.0).
- 13.2 Backend routing as a state machine: the 16-cell matrix
  (4 backends × has-code × has-host) with documented routes.
- 13.3 Opt-in optimisation flags: gas_estimation_mode,
  fast_value_transfer + _acknowledged guard, EIP-2929 pre-warm.
  Default-off; fast path requires explicit ack.
- 13.4 Empirical performance ceiling: dispatcher curve plateaus at
  0.53× CPU at N=32K; bench_kernel hits 1.84× at 11M opcodes.
  Crossover analysis ties to the 1.2M-op threshold from §9.
- 13.5 Four-way parity tightening: 15-test dispatch matrix +
  30 precompile tests (was 19) covering BN256_PAIRING, BLAKE2F,
  POINT_EVALUATION, BLS12-381 (7 ops). Outstanding: 26
  KernelCpuMissing carve-outs on the CPU spec witness.
- 13.6 v0.25.0 roadmap: V2 Metal kernel, async dispatch, full CPU
  interpreter, EIP-2929 wire-up, real GPU CALL via GpuHostBridge.

Conclusion updated to reference v0.24.0 + the closing correctness
gap. Cites verifiable luxcpp/evm commit SHAs throughout.

PDF rebuilt: 231 KiB → 237 KiB (~6 KiB added for §13).
2026-03-03 16:00:00 -08:00

69 lines
2.8 KiB
TeX

\section{Conclusion}
%% ============================================================================
We have shown that the Lux multi-chain admits a uniform DAG consensus
model in which the execution engine's conflict graph \emph{is} the
consensus DAG, yielding a single implementation pattern that covers EVM,
UTXO, ZK-UTXO, DEX, Oracle, Relay, AI, and Service-Node workloads. Safety,
serialisability, and double-spend prevention are proved in Lean~4 and
model-checked in TLA$^+$.
The GPU EVM stack delivered through tags \texttt{v0.22.0},
\texttt{v0.23.0}, and \texttt{v0.24.0} of the \texttt{luxcpp/evm}
repository extends the critical-path argument: ecrecover and keccak
retain their original $32.3\times$ and $7.1\times$ speedups, the Metal
kernel is opcode-complete on Cancun, the CUDA port reproduces the Metal
kernel under a four-way parity test (133 hand-crafted bytecode vectors,
four independent backends, bit-exact agreement on gas/status/output),
and 100\% function coverage on the kernel CPU spec witness ensures every
implemented opcode is exercised on every CI run. Three consensus-relevant
bugs (mulmod carry-loss, unrecognized-opcode gas accounting, MSTORE
offset overflow) were surfaced by the differential-oracle property of
cross-implementation parity and fixed in both kernels. The v0.24.0
correctness release (\S\ref{sec:v024}) contractualises the
dispatcher--consumer seam, closes 8 of 22 Red HIGH findings outright,
and documents the precise performance ceiling --- with five named
priorities for v0.25.0 to clear the bytecode crossover threshold.
The formal development is open-source at \texttt{lux/formal}; the
benchmarks are at \texttt{lux/evm-bench} and \texttt{lux/benchmarks};
the Go implementation is at \texttt{lux/evmgpu} and \texttt{lux/chains};
the GPU EVM kernel and parity test driver are at \texttt{luxcpp/evm}
under tag \texttt{v0.24.0}.
\begin{thebibliography}{9}
\bibitem{blockstm}
R.~Gelashvili, A.~Spiegelman, Z.~Xiang, G.~Danezis, Z.~Li, D.~Malkhi,
Y.~Xia, R.~Zhou.
\emph{Block-STM: Scaling Blockchain Execution by Turning Ordering Curse to
a Performance Blessing}. PPoPP 2023.
\bibitem{mysticeti}
K.~Babel, A.~Chursin, G.~Danezis, L.~Kokoris-Kogias, A.~Sonnino.
\emph{Mysticeti: Reaching the Limits of Latency with Uncertified DAGs}.
arXiv:2310.14821.
\bibitem{monad}
J.~Tang et al.
\emph{Monad: Decoupling Execution and Consensus for Parallel EVM}.
Monad Labs whitepaper, 2024.
\bibitem{pevm}
V.~Yin, M.~Ho.
\emph{pevm: Parallel EVM}.
Risechain whitepaper, 2024.
\bibitem{reth}
Paradigm. \emph{Reth: A Rust Ethereum Execution Layer Client}. 2024.
\bibitem{evmone}
Ipsilon. \emph{evmone: Fast Ethereum Virtual Machine Implementation}.
\bibitem{narwhal}
G.~Danezis, L.~Kokoris-Kogias, A.~Sonnino, A.~Spiegelman.
\emph{Narwhal and Tusk: A DAG-based Mempool and Efficient BFT Consensus}.
EuroSys 2022.
\end{thebibliography}