papers/lp-217-cert-profile-modes/sections/02-modes.tex

\section{The Four Modes}

Each mode defines a REQUIRED leg set. A cert is at mode $M$ if and
only if every leg in $M$'s required set is present and verified.
Modes are totally ordered by subset inclusion within their family
(classical or pure-PQ).

\begin{table}[h]
\centering
\small
\begin{tabular}{@{}lp{3.8cm}p{3.5cm}rr@{}}
\toprule
\textbf{Mode} & \textbf{Required legs} & \textbf{Recommended use} & \textbf{M1 wall} & \textbf{Spark wall} \\
\midrule
\texttt{PQ-off} & BLS & HFT trading; non-financial L2s & \SI{6.9}{\milli\second} & \SI{6.3}{\milli\second} \\
\texttt{PQ-fast} & BLS + Pulsar & Payment rails; gaming; DEX matching & \SI{391}{\milli\second} & \SI{585}{\milli\second} \\
\texttt{PQ-strict} & BLS + Pulsar + Corona & \textbf{Default} -- custody, treasury, regulated securities & \SI{360}{\milli\second} & \SI{595}{\milli\second} \\
\texttt{PQ-heavy} & BLS + Corona + Pulsar + Magnetar & Cross-chain bridges, sovereign anchors & \SI{2.22}{\second} & \SI{1.33}{\second} \\
\bottomrule
\end{tabular}
\caption{Four classical modes with measured wall-clock per signing
round at $N{=}64$ (Quasar 4-scheme matrix bench, 2026-06-04). M1 Max
(10-core darwin/arm64); Spark = NVIDIA GB10 Grace Blackwell
(20-core linux/aarch64). Cert size at $N{=}64$: PQ-off \SI{96}{B},
PQ-fast \SI{33154}{B}, PQ-strict \SI{245634}{B}, PQ-heavy \SI{1.28}{MB}.
Magnetar measured with SLH-DSA-192s -- see \S\ref{sec:magnetar-param}
for the 192f production lock.}
\label{tab:modes}
\end{table}

\subsection{Magnetar parameter set choice -- locked 2026-06-04}
\label{sec:magnetar-param}

SLH-DSA at the 192-bit security level has two parameter sets:
\emph{192s} (small, \SI{16}{KB} sigs, \SI{795}{ms}/sign on Spark) and
\emph{192f} (fast, \SI{35}{KB} sigs, \SI{36}{ms}/sign on Spark). The
matrix bench measured 192s, which is what
\texttt{\textasciitilde/work/lux/threshold/protocols/magnetar/} ships
today. At \SI{795}{ms} per per-validator sign, \emph{no Magnetar-
bearing profile is sub-second at any measured $N$}: PQ-heavy on Spark
at $N{=}64$ blows past \SI{1.33}{s}. Switching to 192f drops the
per-validator sign 22$\times$ to \SI{36}{ms} and restores the
sub-\SI{1}{s} budget. Production deployments configuring PQ-heavy
MUST use 192f.

\subsection{GPU dispatch is unmeasured today}

The matrix bench captured \num{4595} samples of \texttt{nvidia-smi dmon}
during the 2-hour run on Spark: mean \texttt{SM\%} = \SI{11.9}{\percent}
(idle / background only). \textbf{The GPU was not exercised by any
path in this bench.} Two boundary gaps prevent dispatch:

\begin{itemize}[leftmargin=2em,nosep]
\item \texttt{lux-lattice.pc} is not installed on Spark
  (\texttt{find\_package(MLX REQUIRED)} fails on the CUDA target's
  dependency chain), so \texttt{lattice/v7/gpu} doesn't build under
  \texttt{-tags "cgo gpu"}.
\item Magnetar SLH-DSA's CUDA kernel exists and passes KAT tests on
  Spark, but \texttt{lux\_slhdsa\_*} entry points are not yet exposed
  in \texttt{lux/accel/c\_api.h}.
\end{itemize}

GPU numbers cited in earlier LP-203 drafts (PQ-strict
$\approx$\SI{15}{ms}--\SI{50}{ms}) are projections, not measurements.
Until both boundary gaps close, all GPU latency claims in downstream
LPs are tagged ``projected, not measured.''

\subsection{Pure-PQ variants}

\begin{table}[h]
\centering
\small
\begin{tabular}{@{}lp{4cm}p{6cm}@{}}
\toprule
\textbf{Mode} & \textbf{Required legs} & \textbf{Recommended use} \\
\midrule
\texttt{strict-PQ-fast} & Pulsar & Post-PQ-transition L1 with no BLS bridge dependency \\
\texttt{strict-PQ-strict} & Pulsar + Corona & Sovereign PQ-only treasury chain \\
\texttt{strict-PQ-heavy} & Corona + Pulsar + Magnetar & PQ-only long-archival anchor \\
\bottomrule
\end{tabular}
\caption{Pure-PQ variants drop BLS.}
\label{tab:pure-pq}
\end{table}

\subsection{Mode-to-internal mapping}

The v1 codenames remain in the codebase as INTERNAL identifiers
(struct names, package paths, EasyCrypt artifact filenames). The
mode-to-internal map for operator config translation:

\begin{table}[h]
\centering
\small
\begin{tabular}{@{}llll@{}}
\toprule
\textbf{v2 mode} & \textbf{Required legs} & \textbf{Closest v1 codename} & \textbf{Notes} \\
\midrule
\texttt{PQ-off} & BLS & (no v1 name) & v1 implicitly required at least one PQ leg \\
\texttt{PQ-fast} & BLS + Pulsar & v1 ``Pulsar profile'' & Locked by parity audit \\
\texttt{PQ-strict} & BLS + Pulsar + Corona & Aurora & Same lattice diversity; ZK profile-orthogonal in v2 \\
\texttt{PQ-heavy} & BLS + Corona + Pulsar + Magnetar & Polaris & Cross-family maximum \\
\texttt{strict-PQ-*} & PQ legs only & (no v1 name) & New in v2 \\
\bottomrule
\end{tabular}
\caption{v2 mode names mapped to v1 internal codenames.}
\label{tab:v1-v2}
\end{table}

\subsection{Operator config}

One struct -- \texttt{CertPolicy} -- declares the full cert posture
for a chain at genesis. Three previously-scattered knobs
(\texttt{cert\_mode}, \texttt{cert\_timeout\_ms}, and LP-204's per-L1
mode picker) collapse to four fields of one record. LP-202 references
\texttt{CertPolicy.TimeoutMs}; LP-204 attaches a \texttt{CertPolicy}
per chain-VM (default: inherit from parent L1); LP-218 reuses
\texttt{CertPolicy.Mode} as the rollup's inherited cert mode.

\begin{verbatim}
type CertPolicy struct {
    Mode      CertMode     // PQ-off|PQ-fast|PQ-strict|PQ-heavy
    Variant   CertVariant  // Hybrid (with BLS) | Strict (pure PQ)
    TimeoutMs uint32       // max wait for full-mode cert; past TimeoutMs
                          //  LP-202 tier degradation kicks in
    Fallback  CertMode     // tier to settle at if Mode's legs don't
                          //  arrive in TimeoutMs; MUST be <= Mode
}
\end{verbatim}

YAML form:

\begin{verbatim}
quasar:
  cert_policy:
    mode:       PQ-strict
    variant:    hybrid       # hybrid | strict
    timeout_ms: 1000
    fallback:   PQ-fast      # MUST be <= mode
\end{verbatim}

\subsection{Validation rules}

\texttt{CertPolicy} is validated at chain genesis. A chain MUST
refuse to launch with an invalid \texttt{CertPolicy}:

\begin{enumerate}[leftmargin=2em,nosep]
\item \texttt{Fallback <= Mode}. The fallback tier cannot be stronger
  than the target mode; the four modes are subset-ordered.
\item \texttt{Variant == Strict} requires \texttt{Mode >= PQ-fast}.
  Pure-PQ at \texttt{PQ-off} is the empty leg set; refused.
\item \texttt{TimeoutMs >= 2 * expected\_floor\_latency(Mode)}. A
  chain configured at a high \texttt{Mode} with a \texttt{TimeoutMs}
  shorter than the floor latency for that mode will routinely publish
  certs at a lower tier; the validator MUST reject such a
  configuration at genesis (no operationally-useful chain declares
  PQ-heavy and then never satisfies it). Floor latencies at
  Blackwell $N{=}64$:
  PQ-off $\approx$\SI{1}{\milli\second} (min TimeoutMs
  \SI{2}{\milli\second});
  PQ-fast $\approx$\SI{5}{\milli\second} (min \SI{10}{\milli\second});
  PQ-strict $\approx$\SI{50}{\milli\second} (min
  \SI{100}{\milli\second});
  PQ-heavy $\approx$\SI{80}{\milli\second} (min
  \SI{160}{\milli\second}).
\item \texttt{Fallback} as a \texttt{(Mode, Variant)} pair under the
  chain's \texttt{Variant} must itself satisfy rule 2. A chain at
  \texttt{Variant=Strict} cannot fall back to \texttt{PQ-off} (the
  eighth-slot configuration).
\end{enumerate}