mirror of
https://github.com/lux-apps/papers.git
synced 2026-10-30 18:15:21 +00:00
Lux Partners Limited is not the entity behind this work. Every place the old name stood — copyright lines, site footers, brand config, legal and compliance text — now names the company that actually holds it. lux.fund is the exception and is not renamed to match: that site belongs to Lux Assets Management LLC, a different company, and it says so. Two kinds of line were left for a person rather than rewritten. A dated news item recording that Lux Partners Limited was established in the Isle of Man is a historical fact, and renaming it would falsify the record. And a sentence asserting that the company is domiciled in the Isle of Man is a claim about a jurisdiction, not a name — an Inc or an LLC is not an Isle of Man company, so swapping the name there would assert something untrue.
422 lines
14 KiB
TeX
422 lines
14 KiB
TeX
% Options for packages loaded elsewhere
|
|
\PassOptionsToPackage{unicode}{hyperref}
|
|
\PassOptionsToPackage{hyphens}{url}
|
|
\documentclass[
|
|
]{article}
|
|
\usepackage{xcolor}
|
|
\usepackage{amsmath,amssymb}
|
|
\setcounter{secnumdepth}{-\maxdimen} % remove section numbering
|
|
\usepackage{iftex}
|
|
\ifPDFTeX
|
|
\usepackage[T1]{fontenc}
|
|
\usepackage[utf8]{inputenc}
|
|
\usepackage{textcomp} % provide euro and other symbols
|
|
\else % if luatex or xetex
|
|
\usepackage{unicode-math} % this also loads fontspec
|
|
\defaultfontfeatures{Scale=MatchLowercase}
|
|
\defaultfontfeatures[\rmfamily]{Ligatures=TeX,Scale=1}
|
|
\fi
|
|
\usepackage{lmodern}
|
|
\ifPDFTeX\else
|
|
% xetex/luatex font selection
|
|
\fi
|
|
% Use upquote if available, for straight quotes in verbatim environments
|
|
\IfFileExists{upquote.sty}{\usepackage{upquote}}{}
|
|
\IfFileExists{microtype.sty}{% use microtype if available
|
|
\usepackage[]{microtype}
|
|
\UseMicrotypeSet[protrusion]{basicmath} % disable protrusion for tt fonts
|
|
}{}
|
|
\makeatletter
|
|
\@ifundefined{KOMAClassName}{% if non-KOMA class
|
|
\IfFileExists{parskip.sty}{%
|
|
\usepackage{parskip}
|
|
}{% else
|
|
\setlength{\parindent}{0pt}
|
|
\setlength{\parskip}{6pt plus 2pt minus 1pt}}
|
|
}{% if KOMA class
|
|
\KOMAoptions{parskip=half}}
|
|
\makeatother
|
|
\usepackage{longtable,booktabs,array}
|
|
\newcounter{none} % for unnumbered tables
|
|
\usepackage{calc} % for calculating minipage widths
|
|
% Correct order of tables after \paragraph or \subparagraph
|
|
\usepackage{etoolbox}
|
|
\makeatletter
|
|
\patchcmd\longtable{\par}{\if@noskipsec\mbox{}\fi\par}{}{}
|
|
\makeatother
|
|
% Allow footnotes in longtable head/foot
|
|
\IfFileExists{footnotehyper.sty}{\usepackage{footnotehyper}}{\usepackage{footnote}}
|
|
\makesavenoteenv{longtable}
|
|
\setlength{\emergencystretch}{3em} % prevent overfull lines
|
|
\providecommand{\tightlist}{%
|
|
\setlength{\itemsep}{0pt}\setlength{\parskip}{0pt}}
|
|
\usepackage{bookmark}
|
|
\IfFileExists{xurl.sty}{\usepackage{xurl}}{} % add URL line breaks if available
|
|
\urlstyle{same}
|
|
\hypersetup{
|
|
hidelinks,
|
|
pdfcreator={LaTeX via pandoc}}
|
|
|
|
\author{}
|
|
\date{}
|
|
|
|
\begin{document}
|
|
|
|
\section{QuasarSTM 4.0 Activation ---
|
|
2026-02-14}\label{quasarstm-40-activation--2026-02-14}
|
|
|
|
\textbf{Audience:} Lux validators, application developers, ecosystem
|
|
partners \textbf{Activation:} 2026-02-14, 17:00 UTC \textbf{Spec:}
|
|
LP-135 (production spec) and LP-010-quasar-stm-4 (paper)
|
|
|
|
\begin{center}\rule{0.5\linewidth}{0.5pt}\end{center}
|
|
|
|
\subsection{TL;DR}\label{tldr}
|
|
|
|
QuasarSTM 4.0 is live on Lux mainnet as of 2026-02-14, 17:00 UTC. It is
|
|
the production version of the GPU-native ordered MVCC execution fabric
|
|
that 3.0 (2025-12-25) shipped as the substrate. Every gap 3.0 left open
|
|
is now closed:
|
|
|
|
\begin{itemize}
|
|
\tightlist
|
|
\item
|
|
\textbf{Real cryptography.} BLS12-381, Pulsar Module-LWE, and Groth16
|
|
verifiers run on-device. The HMAC-keccak placeholders that 3.0 shipped
|
|
at launch are gone from production.
|
|
\item
|
|
\textbf{Full EVM.} All 175 opcodes including the entire
|
|
\texttt{CALL}/\texttt{CREATE} family, \texttt{LOG}n,
|
|
\texttt{EXTCODE*}, \texttt{RETURNDATA*},
|
|
\texttt{TLOAD}/\texttt{TSTORE}, \texttt{MCOPY},
|
|
\texttt{BLOBHASH}/\texttt{BLOBBASEFEE}. Journaled SSTORE with EIP-2929
|
|
cold/warm and EIP-3529 refunds.
|
|
\item
|
|
\textbf{Production validation.} Tier 1 lane-clock fast path, Tier 3
|
|
semantic reducer commit at horizon time, ConflictSpec ABI, NEMO
|
|
LaneClass, dynamic per-lane versioning mode.
|
|
\item
|
|
\textbf{Production execution.} Fiber checkpoints per
|
|
\texttt{CALL}-frame, commit horizon over Nova prefixes / Nebula causal
|
|
cuts, MVCC GC bound to the horizon, Motor-style VersionBlock layout (8
|
|
inline versions per key).
|
|
\item
|
|
\textbf{Production GPU pipeline.} 16 service drains (the 12 from 3.0
|
|
plus AdaptiveSchedule, BridgeAttest, MarketAuction, FiberCheckpoint
|
|
--- A/B/M/F).
|
|
\item
|
|
\textbf{Production rigor.} CSMV-style commit-server scaffold, formally
|
|
specified CPU reference, cross-backend determinism harness gated in CI
|
|
at \textgreater= 80\% line coverage on Apple Metal and NVIDIA CUDA.
|
|
\end{itemize}
|
|
|
|
The wire format is unchanged. The public host API
|
|
(\texttt{QuasarGPUEngine}) is unchanged. The 3.0 invariants ---
|
|
determinism, ordered serializability, repair monotonicity, horizon
|
|
safety --- are preserved verbatim.
|
|
|
|
\begin{center}\rule{0.5\linewidth}{0.5pt}\end{center}
|
|
|
|
\subsection{What changed from 3.0}\label{what-changed-from-30}
|
|
|
|
{\def\LTcaptype{none} % do not increment counter
|
|
\begin{longtable}[]{@{}lll@{}}
|
|
\toprule\noalign{}
|
|
Topic & 3.0 (2025-12-25) & 4.0 (2026-02-14) \\
|
|
\midrule\noalign{}
|
|
\endhead
|
|
\bottomrule\noalign{}
|
|
\endlastfoot
|
|
EVM coverage & 118 opcodes (no \texttt{CALL}, no \texttt{CREATE}, no
|
|
\texttt{LOG}, no \texttt{EXTCODE*}) & 175 opcodes --- full coverage \\
|
|
SSTORE & linear write-through & journaled with EIP-2929 / EIP-3529 \\
|
|
BLS verifier & HMAC-keccak placeholder & real BLS12-381 pairing
|
|
on-device \\
|
|
Pulsar verifier & HMAC-keccak placeholder & real Module-LWE share
|
|
verify \\
|
|
Groth16 verifier & HMAC-keccak placeholder & real Groth16 over
|
|
BLS12-381 \\
|
|
Tier 1 lane-clock & sketch & production fast path \\
|
|
Tier 3 semantic reducers & enum stub & production reducer commit \\
|
|
ConflictSpec ABI & not implemented & production (Static / ABI /
|
|
Historical / UserDeclared / Precompile / Learned) \\
|
|
LaneClass & not implemented & NEMO classes (Owned / Shared / HotShared /
|
|
Commutative / Unknown) \\
|
|
Versioning mode & always-on multi-version & dynamic per-lane
|
|
(SingleVersionFast / MultiVersion / Reducer / Serialized) \\
|
|
Fiber checkpoints & not implemented & production, per CALL-frame \\
|
|
Commit horizon & sketch & production (Nova prefix / Nebula causal
|
|
cut) \\
|
|
MVCC GC & not implemented & production, horizon-driven \\
|
|
Motor VersionBlock & not implemented & production (8 inline versions per
|
|
key) \\
|
|
GPU services & 12 & 16 (12 + A/B/M/F) \\
|
|
CSMV commit server & not implemented & scaffold landed, single-GPU
|
|
production \\
|
|
Formal CPU reference & not implemented & small executable semantics +
|
|
differential fuzzer \\
|
|
Cross-backend determinism & per-backend tests & single harness,
|
|
byte-identical roots Metal == CUDA \\
|
|
Coverage & informal & \textgreater= 80\% line coverage on both backends,
|
|
gated in CI \\
|
|
\end{longtable}
|
|
}
|
|
|
|
The 4.0 deliverable train shipped over six weeks (2026-01-15 through
|
|
2026-02-07 spec freeze), as v0.41 through v0.49 milestones in
|
|
\texttt{cevm}:
|
|
|
|
{\def\LTcaptype{none} % do not increment counter
|
|
\begin{longtable}[]{@{}ll@{}}
|
|
\toprule\noalign{}
|
|
Milestone & Theme \\
|
|
\midrule\noalign{}
|
|
\endhead
|
|
\bottomrule\noalign{}
|
|
\endlastfoot
|
|
v0.41 & CUDA backend mirror --- persistent CTAs,
|
|
\texttt{\_\_threadfence}, layout-byte-identical with Metal \\
|
|
v0.42 & Cert-subject hardening, \texttt{KnownTotalOrder}, separated mode
|
|
roots \\
|
|
v0.43 & ConflictSpec ABI + LaneClass + dynamic VersioningMode +
|
|
AdaptiveSchedule drain \\
|
|
v0.44 & Real BLS12-381 pairing kernel \\
|
|
v0.45 & Real Pulsar Module-LWE + real Groth16 verifiers \\
|
|
v0.46 & Full EVM coverage + journaled SSTORE + EIP-2929/3529 + Tier 3
|
|
reducers \\
|
|
v0.47 & Predictive scheduling + Chiron-style hint roots \\
|
|
v0.48 & Motor VersionBlock + A/B/M/F drains + fiber checkpoints \\
|
|
v0.49 & CSMV commit-server scaffold + formal CPU reference +
|
|
cross-backend determinism harness \\
|
|
\end{longtable}
|
|
}
|
|
|
|
\begin{center}\rule{0.5\linewidth}{0.5pt}\end{center}
|
|
|
|
\subsection{Performance}\label{performance}
|
|
|
|
Apple M1 Max, 1024-tx end-to-end stress on the regulated-DEX workload
|
|
mix (50\% per-account orders, 20\% cancels, 15\% fee accumulation, 10\%
|
|
balance deltas, 5\% admin):
|
|
|
|
{\def\LTcaptype{none} % do not increment counter
|
|
\begin{longtable}[]{@{}lll@{}}
|
|
\toprule\noalign{}
|
|
Metric & 3.0 & 4.0 \\
|
|
\midrule\noalign{}
|
|
\endhead
|
|
\bottomrule\noalign{}
|
|
\endlastfoot
|
|
Wallclock (median) & 150 ms & \textbf{108 ms} \\
|
|
Wave ticks & 8 & \textbf{6} \\
|
|
Repair amplification & \textasciitilde0.97 &
|
|
\textbf{\textasciitilde0.42} \\
|
|
EVM coverage & 118/175 & \textbf{175/175} \\
|
|
BLS verify (per cert) & 9 us (HMAC-keccak) & 46 us (real BLS12-381) \\
|
|
Pulsar verify (per share) & 7 us (placeholder) & 38 us (real
|
|
Module-LWE) \\
|
|
Groth16 verify (per cert) & 8 us (placeholder) & 62 us (real Groth16) \\
|
|
\end{longtable}
|
|
}
|
|
|
|
The 28\% wallclock improvement is not because 4.0 makes the same
|
|
operations cheaper --- real cryptography is 4-7x more expensive than the
|
|
HMAC-keccak placeholders. The wins come from Tier 1 lane-clock
|
|
fast-pathing \textgreater80\% of commits without an MVCC chain walk and
|
|
from Tier 3 reducers absorbing the contended writes (fee accumulators,
|
|
order-book level updates, balance deltas) that 3.0 turned into same-key
|
|
SSTORE conflicts.
|
|
|
|
The canonical Block-STM hot-key test (16 same-key contending
|
|
transactions producing 120 conflicts -\textgreater{} 120 repairs
|
|
-\textgreater{} 16 commits) is unchanged in 4.0 and is the determinism
|
|
harness\textquotesingle s headline regression. Same input
|
|
-\textgreater{} byte-identical roots on Metal and CUDA.
|
|
|
|
NVIDIA H100 cert costs for reference:
|
|
|
|
\begin{itemize}
|
|
\tightlist
|
|
\item
|
|
BLS12-381 aggregate verify: 31 us
|
|
\item
|
|
Pulsar share verify: 24 us
|
|
\item
|
|
Groth16 verify: 42 us
|
|
\end{itemize}
|
|
|
|
\begin{center}\rule{0.5\linewidth}{0.5pt}\end{center}
|
|
|
|
\subsection{Migration}\label{migration}
|
|
|
|
\subsubsection{Validators}\label{validators}
|
|
|
|
\begin{enumerate}
|
|
\def\labelenumi{\arabic{enumi}.}
|
|
\tightlist
|
|
\item
|
|
Upgrade \texttt{cevm} to a release built from \texttt{cevm}
|
|
\textgreater= v0.49.
|
|
\item
|
|
Re-roll the Quasar consensus engine (LP-020) with
|
|
\texttt{execution\_version=4} support.
|
|
\item
|
|
Verify the cert verifier table at engine creation. On production
|
|
networks the verifier table must be the production table; the
|
|
\texttt{EVM\_DEV\_HMAC\_VERIFIER} flag is rejected at startup.
|
|
\item
|
|
Run \texttt{cevm-genesis-replay} against the latest 3.0 round to
|
|
confirm byte-identical roots before publishing.
|
|
\item
|
|
Publish on the validator gossip channel and join 4.0 round production.
|
|
\end{enumerate}
|
|
|
|
There is no on-disk state migration. 4.0 reads 3.0 round artifacts as
|
|
historical 3.0 rounds. New rounds emitted post-activation are 4.0
|
|
rounds.
|
|
|
|
Validators that remain on 3.0 after 17:00 UTC on 2026-02-14 are rejected
|
|
by the consensus engine and must upgrade to participate.
|
|
|
|
\subsubsection{Application developers}\label{application-developers}
|
|
|
|
\begin{enumerate}
|
|
\def\labelenumi{\arabic{enumi}.}
|
|
\tightlist
|
|
\item
|
|
Recompile contracts against the 4.0 EVM (175 opcodes). Contracts that
|
|
compiled against 3.0 (118 opcodes) are forward-compatible.
|
|
\item
|
|
Adopt the Solidity \texttt{lane\_class} attribute on hot state
|
|
variables to opt into the AdaptiveSchedule fast path. Without the
|
|
attribute, the predictor and historical sources still apply; the
|
|
attribute adds a higher-priority \texttt{Static} source.
|
|
\item
|
|
For DEX-style hot lanes (fee accumulators, order-book levels), adopt
|
|
the \texttt{Commutative} class to route writes through Tier 3
|
|
reducers. The repair-amplification reduction comes from this change.
|
|
\item
|
|
Re-run integration tests against the cross-backend determinism
|
|
harness; contracts that produce identical roots on Metal and CUDA are
|
|
4.0-clean.
|
|
\end{enumerate}
|
|
|
|
\subsubsection{Ecosystem partners}\label{ecosystem-partners}
|
|
|
|
\begin{itemize}
|
|
\tightlist
|
|
\item
|
|
Bridge operators: BridgeAttest is a new GPU drain with a stable ABI;
|
|
existing bridge messages route through it without re-signing.
|
|
\item
|
|
DEX operators: MarketAuction runs batched matching on-device. The
|
|
\texttt{AuctionMatch} reducer entry replaces per-trade SSTORE
|
|
conflicts; existing order/cancel/fill flows are forward-compatible.
|
|
\item
|
|
Wallet integrators: no API changes. The wire-format is unchanged.
|
|
\end{itemize}
|
|
|
|
\begin{center}\rule{0.5\linewidth}{0.5pt}\end{center}
|
|
|
|
\subsection{Backward compatibility}\label{backward-compatibility}
|
|
|
|
{\def\LTcaptype{none} % do not increment counter
|
|
\begin{longtable}[]{@{}ll@{}}
|
|
\toprule\noalign{}
|
|
Component & 3.0 -\textgreater{} 4.0 compatibility \\
|
|
\midrule\noalign{}
|
|
\endhead
|
|
\bottomrule\noalign{}
|
|
\endlastfoot
|
|
Wire format (Quasar consensus) & unchanged; LP-020 cert-lane structure
|
|
preserved \\
|
|
\texttt{QuasarRoundDescriptor} / \texttt{QuasarRoundResult} & unchanged
|
|
byte-for-byte; \texttt{hint\_roots} ride in the existing \texttt{roots}
|
|
slot \\
|
|
Public host API (\texttt{QuasarGPUEngine}) & unchanged \\
|
|
\texttt{ServiceId} enum & extended (\texttt{Count} 12 -\textgreater{}
|
|
16); old IDs unchanged \\
|
|
Layout types & \texttt{VersionBlock.MAX\_INLINE\_VERSIONS} 4
|
|
-\textgreater{} 8 (forward-compat: 4.0 reads 3.0 blocks) \\
|
|
HMAC-keccak verifier path & gated behind
|
|
\texttt{EVM\_DEV\_HMAC\_VERIFIER=1} for legacy test vectors \\
|
|
Validators running 3.0 after 2026-02-14 & rejected at the consensus
|
|
engine --- must upgrade \\
|
|
\end{longtable}
|
|
}
|
|
|
|
There is no on-disk state migration. Pre-activation rounds remain valid
|
|
as historical 3.0 rounds; the activation height is the partition.
|
|
|
|
\begin{center}\rule{0.5\linewidth}{0.5pt}\end{center}
|
|
|
|
\subsection{Known limitations (out of scope for
|
|
4.0)}\label{known-limitations-out-of-scope-for-40}
|
|
|
|
The following are explicitly \emph{not} in 4.0 and are tracked under
|
|
QuasarSTM 5.0 research:
|
|
|
|
\begin{itemize}
|
|
\tightlist
|
|
\item
|
|
Multi-GPU MVCC over RDMA. 4.0 ships the Motor-style VersionBlock
|
|
layout (which is RDMA-friendly) and the CSMV commit-server scaffold
|
|
(which is the multi-GPU pattern), but the production deployment is
|
|
single-GPU per validator.
|
|
\item
|
|
Operation-window validation for pathological long transactions.
|
|
\item
|
|
Formal proof of kernel refinement against the abstract semantics. 4.0
|
|
ships the abstract semantics in \texttt{lib/consensus/quasar/spec/}
|
|
(executable C++ + Lean 4 stubs) and the differential fuzzer; the
|
|
refinement proof is 5.0.
|
|
\item
|
|
Distributed CSMV commit server (4.0 ships single-GPU).
|
|
\item
|
|
800G InfiniBand / GPUDirect cell topologies.
|
|
\end{itemize}
|
|
|
|
These are not on the 4.0 production path. They will land in 5.0 on a
|
|
schedule TBD post-activation review.
|
|
|
|
\begin{center}\rule{0.5\linewidth}{0.5pt}\end{center}
|
|
|
|
\subsection{References}\label{references}
|
|
|
|
\begin{itemize}
|
|
\tightlist
|
|
\item
|
|
\textbf{LP-135} --- QuasarSTM 4.0 production spec
|
|
\item
|
|
\textbf{LP-010-quasar-stm-4} --- QuasarSTM 4.0 paper (this
|
|
announcement)
|
|
\item
|
|
\textbf{LP-010} --- QuasarSTM 3.0 substrate (2025-12-25)
|
|
\item
|
|
\textbf{LP-020} --- Quasar Consensus 3.0 (cert lanes, P/Q/Z)
|
|
\item
|
|
\textbf{LP-132} --- QuasarGPU Execution Adapter (16 services
|
|
post-2026-02-14)
|
|
\item
|
|
\textbf{LP-009} --- GPU-Native EVM
|
|
\item
|
|
\textbf{LP-070} --- ML-DSA
|
|
\item
|
|
\textbf{LP-073} --- Pulsar
|
|
\item
|
|
\textbf{LP-075} --- BLS
|
|
\end{itemize}
|
|
|
|
For questions:
|
|
\href{mailto:[email protected]}{\nolinkurl{[email protected]}} For
|
|
validator support:
|
|
\href{mailto:[email protected]}{\nolinkurl{[email protected]}} For
|
|
application integration:
|
|
\href{mailto:[email protected]}{\nolinkurl{[email protected]}}
|
|
|
|
\begin{center}\rule{0.5\linewidth}{0.5pt}\end{center}
|
|
|
|
Copyright (C) 2026, Lux Industries Inc. All rights reserved.
|
|
|
|
\end{document}
|