mirror of
https://github.com/lux-apps/papers.git
synced 2026-11-03 05:07:46 +00:00
zeekay
98343d238e
Quasar Horizon (consensus composition over Pulsar / ML-DSA / BLS lanes
under Prism + LSS lifecycle), Warp 2.0 (Pulse-backed cross-chain envelope
with forward-only v1->v2 migration and lattigo ReadUint64Slice DoS fix),
benchmark report (microbench, reshare, activation, Warp 2.0 verify, Prism
composition, GPU headroom, cross-WAN latency), and NIST IR 8214C-style
submission package (spec, parameter sets, domain separation, serialization,
reference-impl pointers, evaluation report, security analysis, side-channel
considerations).
All four deliverables compile cleanly via pdflatex+bibtex and reference
the canonical proof bucket files at proofs/{definitions,pulsar,lss,quasar}.
No duplication of proof bodies; cite by file path.
102 lines
4.0 KiB
TeX
102 lines
4.0 KiB
TeX
\section{ConflictSpec ABI}
|
|
|
|
\label{sec:conflictspec}
|
|
%% ============================================================================
|
|
|
|
\subsection{From blind optimism to fallback-safe oracle}
|
|
|
|
Block-STM is blind optimism by default---it executes optimistically and
|
|
discovers contention retroactively. The AFT~2025 conflict-specifications
|
|
work~\cite{aft2025conflictspec} reports significant speedups when
|
|
declared access information is available before execution, with
|
|
Block-STM as the correctness backstop. QuasarSTM~4.0 ships this as the
|
|
production scheduler oracle.
|
|
|
|
\subsection{Record}
|
|
|
|
\begin{lstlisting}[language=C++]
|
|
struct ConflictSpec {
|
|
uint32_t tx_id;
|
|
uint32_t read_lane_offset;
|
|
uint16_t read_lane_count;
|
|
uint32_t write_lane_offset;
|
|
uint16_t write_lane_count;
|
|
uint32_t commutative_lane_offset;
|
|
uint16_t commutative_lane_count;
|
|
uint8_t confidence; // 0..255
|
|
uint8_t source; // ConflictSpecSource
|
|
};
|
|
|
|
enum class ConflictSpecSource : uint8_t {
|
|
Static = 0, // compile-time ABI declaration
|
|
ABI = 1, // EVM ABI hint (Solidity attribute)
|
|
Historical = 2, // observed past traces
|
|
UserDeclared = 3, // signed declaration from sender
|
|
Precompile = 4, // precompile self-declaration
|
|
Learned = 5, // device-resident predictor (ForeSight-style)
|
|
};
|
|
\end{lstlisting}
|
|
|
|
The \texttt{(read|write|commutative)\_lane\_offset, \_count} pairs
|
|
index into a per-round dense \texttt{lane\_id[]} array stored in the
|
|
\texttt{items\_arena} of the AdaptiveSchedule ring (\S\ref{sec:abmf}).
|
|
|
|
\subsection{Source priority}
|
|
|
|
The scheduler consults sources in fixed precedence:
|
|
|
|
\[
|
|
\textbf{Static} \succ \textbf{ABI} \succ \textbf{UserDeclared} \succ \textbf{Precompile} \succ \textbf{Historical} \succ \textbf{Learned} \succ \textbf{fallback Block-STM}
|
|
\]
|
|
|
|
Concretely, the AdaptiveSchedule drain produces the highest-priority
|
|
\texttt{ConflictSpec} available for each transaction; if none is
|
|
available, the transaction enters Block-STM speculative execution
|
|
exactly as in 3.0~\cite{lp010-quasarstm-3}.
|
|
|
|
\subsection{Why this is correctness-preserving}
|
|
|
|
ConflictSpec is \emph{advice}, not authority. The validation tiers
|
|
(Tier~1 lane-clock, Tier~2 key-MVCC, Tier~3 semantic reducer) execute
|
|
unchanged. A mis-declared \texttt{ConflictSpec} loses speed (the
|
|
speculator schedules suboptimally) but cannot break safety: the
|
|
underlying validators detect the missed read or write and trigger
|
|
repair~\cite{aft2025conflictspec}.
|
|
|
|
\begin{invariant}[ConflictSpec safety]
|
|
For any transaction~$t$ scheduled with $\mathit{ConflictSpec}_t = c$,
|
|
the commit of~$t$ requires Tier~1 / 2 / 3 validation against $t$'s
|
|
\emph{actual} read/write set, not~$c$.
|
|
\end{invariant}
|
|
|
|
In practice the predictor~\cite{foresight2025} and historical-trace
|
|
sources are tuned to over-declare reads and writes (false-positive bias)
|
|
to avoid the missed-conflict path. The Learned source's
|
|
\texttt{confidence} field is published in the round result; validators
|
|
that consistently fall back to speculative STM emit a hint root
|
|
(\S\ref{sec:abmf}) that other validators can ingest as a
|
|
\texttt{Historical} source for subsequent rounds~\cite{chiron2025}.
|
|
|
|
\subsection{Predictor structure}
|
|
|
|
\begin{lstlisting}[language=C++]
|
|
struct LanePredictorEntry {
|
|
Hash code_hash;
|
|
uint32_t selector; // 4-byte function selector
|
|
Hash caller_class;
|
|
Hash market_id;
|
|
Hash predicted_lanes_root; // commitment to declared lanes
|
|
uint16_t confidence; // 0..1000
|
|
uint16_t observed_conflict_rate;
|
|
};
|
|
\end{lstlisting}
|
|
|
|
The predictor is device-resident and updated at end-of-round from the
|
|
observed conflict matrix. Updates are deterministic in the round
|
|
inputs (canonical-order trace of conflicts and commits), so two
|
|
validators running 4.0 against the same canonical order maintain
|
|
byte-identical predictor state. This is required for cross-validator
|
|
hint-root agreement~\cite{chiron2025}.
|
|
|
|
%% ============================================================================
|