Files
zeekayandHanzo Dev 3c29dd6a3e lp-200-zap-stack: codec kill complete + real measured block-level numbers
- 'What is NOT yet decomplected' → 'Decomplection status': the reflection tx
  codec is fully removed (Tx.Sign no longer routes through codec.Manager; every
  VM native ZAP; residual luxfi/codec is only the leaf wrappers packer + Warp
  codec). Kept the orthogonal p2p transport-migration items.
- Performance table: replaced the old estimates (3.6x-13.4x, ~5x/~3x/~10x) with
  fresh measured head-to-head vs the reflection codec (avalanchego master runs):
  parse tx 73x, validate 1000-tx block 70x, produce block 5.7x, memory held for
  a live block 114x (17KB vs 1990KB), 51x fewer allocs. M1 Max, count=5 medians.
- Abstract: same real numbers replace the estimate range.
Compiles clean.

Co-authored-by: Hanzo Dev <[email protected]>
2026-07-14 18:37:34 -07:00

79 lines
3.7 KiB
TeX

\section{Composition with Sibling LPs}
\label{sec:composition}
The ZAP Stack is the umbrella. Every other architectural LP in the
new network is one layer of this stack or one orthogonal extension of
it. The composition table below is the canonical map.
\begin{table}[h]
\centering
\small
\begin{tabular}{@{}llp{8cm}@{}}
\toprule
\textbf{LP} & \textbf{Layer} & \textbf{Composition} \\
\midrule
LP-022 & 0 & ZAP wire frame format; the bottom layer all other LPs read \\
LP-023 & 2 & P-chain (platformvm) wire ZAP-native; \texttt{StakerKind} discriminator \\
LP-077 & 4 & Round-digest binding; the input to the consensus transcript \\
LP-079 & 9 & Q-Chain finality blocks; where cert outputs land \\
LP-181 & 9 & Magnetar threshold SLH-DSA; PQ-heavy cert leg \\
LP-182 & 9 & QuasarCert wire format; composes legs as opaque ZAP byte strings \\
LP-184 & 2 & X-chain (xvm) wire ZAP-native \\
LP-185 & 2 & EVM internal wire ZAP; RLP preserved only at RPC boundary \\
LP-186 & 2 & chains-VM family wire ZAP (twelve VMs) \\
LP-189 & 2 & platformvm state ZAP (pending numbering) \\
LP-201 & 10 & QUIC + ZAP + mDNS + DHT P2P; replaces \texttt{luxfi/p2p} \\
LP-202 & 1, 7 & Pipelining + atomic unwind primitives; MVCC overlays on ZapDB \\
LP-203 & 6 & GPU-native verify; Blackwell sm\_120 kernel for the validator layer \\
LP-204 & --- & Network of blockchains; tier-1 / tier-2 / tier-3 hierarchy \\
LP-205 & 9 & HotStuff-pipelined Quasar; 3 in-flight blocks across the Select/Build/Sign stages \\
LP-208 & 0, 9 & DAG mempool; per-validator header DAG over the ZAP wire \\
LP-209 & 9 & Mysticeti-style total-order on the LP-208 DAG \\
LP-210 & 7 & Block-STM optimistic parallel execution on ZapDB MVCC \\
LP-211 & 9 & Cross-shard atomic Polaris cert via parent-L1 QuasarCert \\
LP-217 & 9 & Cert profile modes -- operator-facing naming \\
LP-218 & 0, 2, 7 & ZAP P3Q PQ rollup; rollup tx envelopes at schema range 0xE0--0xEF \\
\bottomrule
\end{tabular}
\caption{Composition of every architectural LP onto the ten-layer
model.}
\label{tab:composition}
\end{table}
\subsection{Reading the table}
Every cell in the ``Layer'' column points to one row of
Table~\ref{tab:tenlayer}. An LP that spans multiple layers (LP-202
pipelining touches ZapDB layer~1 and the executor layer~7; LP-218
rollup touches the frame layer~0, the schema layer~2, and the
executor layer~7) is honestly multi-layer; we do not pretend
otherwise.
\subsection{Decomplection status}
The reflection transaction codec is fully removed. \texttt{Tx.Sign} in
\texttt{platformvm} and \texttt{xvm} no longer routes through
\texttt{codec.Manager}: the P-Chain and X-Chain transaction \emph{is}
the ZAP buffer (struct $=$ wire), signed over $\mathrm{hash}(\text{unsignedBytes})$.
Zero \texttt{codec.Manager} usages remain in node production, the
reflection dispatcher (\texttt{pcodecs}) is deleted, and every VM's
tx/block wire is native ZAP (P, X, and \texttt{dexvm}, \texttt{mpcvm},
\texttt{zkvm}, \texttt{bridgevm}, \texttt{schain}, \texttt{identityvm},
\texttt{graphvm}). The remaining \texttt{luxfi/codec} dependency is the
leaf \texttt{codec/wrappers} byte-packer used by peer-handshake framing
plus Warp's own cross-chain message codec --- separate leaf subsystems,
not the reflection tx codec.
The residual transport migration is orthogonal to the wire kill:
\begin{itemize}[leftmargin=2em,nosep]
\item \texttt{luxfi/p2p} still imported in eleven files. LP-201
replaces it with QUIC + ZAP + mDNS + DHT. The replacement is in
flight, not landed.
\item \texttt{service/info} still speaks the legacy jsonrpc envelope on
top of \texttt{luxfi/p2p}; replaced when LP-201 lands.
\end{itemize}
Neither mutates the architecture. The architecture is the ten-layer
model; these are transport migrations, not codec ones.