papers/zoo-poai-consensus/sections/00-status-summary.tex

\section{What Is Built vs.\ What Is Designed}
\label{sec:status-summary}

This paper separates three things that earlier drafts blurred together: a proved
verification primitive, reference implementations and adversarial fixtures, and a live
consensus path. The first two exist. The third is an activation-gated integration target.

\paragraph{The honest one-line summary.} Exact-integer Freivalds verification, transcript
construction, graph fixtures, cross-language encodings, and EVM enforcement prototypes are
built and tested. A normal production model does not yet emit the canonical transcript on
every serving path, and the Lux A-Chain does not yet admit the job, finalize the proof,
consume the nullifier, and authorize the mint through this system. Until those wires are
complete, the code demonstrates the construction; it is not evidence of a live PoAI mint.

\begin{table}[h]
\centering
\caption{Current implementation boundary. ``Reference'' means executable and tested, but
not yet the active A-Chain consensus path.}
\label{tab:status-summary}
\small
\begin{tabular}{@{}p{7.4cm}p{2.4cm}p{3.8cm}@{}}
\toprule
Component & Status & Evidence / boundary \\
\midrule
Exact-integer Freivalds verifier and challenge derivation
  & \textbf{built}, tested & Rust and Go fixtures \\
Keccak-Merkle activation transcript and sampled openings
  & \textbf{built}, tested & Reference prover/challenger \\
Proof-bearing linear/graph fixtures for dense, attention, MoE, training
  & \textbf{built}, tested & Coverage fixtures, not a complete Zen serving path \\
Solidity witness, fraud, receipt, and fail-closed gate surfaces
  & \textbf{reference}, tested & EVM prototypes; not A-Chain consensus \\
Canonical deterministic kernel profile and transcript schema
  & \textbf{specified} & Must be versioned and activated by A-Chain policy \\
Normal production-model transcript emission and data availability
  & \textbf{integration} & Not yet complete \\
A-Chain admission, finalized-proof gate, nullifier, metering, and mint authority
  & \textbf{integration} & Sole consensus authority; not yet live \\
Z-Chain P3Q settlement of finalized receipt transitions
  & \textbf{activation-gated} & Shadow/replay first; no mint authority in shadow mode \\
RepOps deterministic floating-point profile and long-tail model coverage
  & \textbf{designed} & Quantized exact-integer profile comes first \\
\bottomrule
\end{tabular}
\end{table}

The test suites establish useful facts: honest exact-integer openings verify; tampered
matmuls are detected with the stated Freivalds error; commitment encodings can be held
stable across implementations; and reference consequence gates can fail closed. They do
not establish production data availability, economic metering, A-Chain finality, or live
P3Q settlement. Cross-language parity is therefore a release-conformance requirement,
not a consensus proof. Likewise, staged activation is operational safety: shadow evidence
may be compared and replayed, but it cannot authorize issuance.

Zoo does not run a second PoAI consensus. Zoo applications can originate useful jobs and
consume finalized receipts. A-Chain alone decides whether useful work is eligible to mine
\textsc{AI}; Z-Chain proves and settles the finalized state transition; Lux Quasar supplies
post-quantum finality; destination chains consume receipt-only, exactly-once payouts.