Architecture

Layer 1

Ingestion and scoring

The ingestion layer maintains a real-time event stream over Base. The stack consists of:

• An Erigon archive node with a custom eth_subscribe filter for the deployment factories of Clanker, Flaunch, Bankr, and Zora.
• A Kafka event bus partitioned by launchpad, with a 7-day retention window for replay and backfill.
• A Postgres-backed enrichment service that joins each new deployment against indexed historical state within 200ms p95.
• A Farcaster Hub subscriber consuming the cast firehose, filtered for token mentions and tags.

Inputs

On-chain signals

• Liquidity lock status and depth (Uniswap v3/v4 pool state, lock contract verification).
• Holder concentration: top-10 share, Gini coefficient, Herfindahl-Hirschman index.
• Dev wallet history: prior deployments via the same EOA, prior rugs by graph traversal of funding paths, ENS / Farcaster ID linkage.
• Contract bytecode pattern matching against known templates and bad actors using a Bloom-filter index of malicious opcodes.

Inputs

Off-chain signals

• Farcaster cast volume and engagement velocity in the first 60 seconds after deploy.
• Casting account's Neynar reputation score and historical hit-rate.
• X mentions weighted by follower-quality score.

The scoring engine outputs a 0–100 composite plus a discrete set of risk flags. See the AI Scoring page for the formal weighted blend.

Layer 2

Privacy primitives

The privacy layer composes four production-grade primitives. Each is detailed in the Privacy section.

• Semaphore over BN254 for zero-knowledge subscription gating.
• XMTP (Signal Protocol) for end-to-end encrypted signal delivery.
• Phala-orchestrated SGX / SEV-SNP enclaves for confidential AI inference.
• Railgun shielded UTXOs for trade-attribution unlinkability.

Layer 3

Delivery and action

Delivery is multi-channel: XMTP inbox, zk-gated Telegram bot, and optional encrypted Farcaster DM. Builders with a premium tier credential receive an authenticated API stream over WebSocket, with rate limits enforced via per-credential token buckets.

The action layer is optional. Users can integrate SHADE as a pre-trade risk check on Bankr; a tagged trade is intercepted, scored, and either confirmed or warned against before execution. Users may also route execution through Railgun or through SHADE's fresh-wallet rotation service.

End-to-end

Protocol flow

// One-time enrollment
wallet w holds >= T $SHADE
commit = Poseidon(nullifier, trapdoor)
submit commit -> Merkle tree T on-chain

// Per session
proof pi = Prove(T, w, epoch)
submit pi -> SHADE; receive session token sigma
subscribe to XMTP topic keyed by sigma

// On each new launch event e
score(e); encrypt under user XMTP key; push to inbox

// On user query q for token tau
encrypt q under enclave key epk
receive encrypted result r

// On user action (buy)
deposit v ETH -> Railgun from w
wait Delta ~ stochastic distribution
withdraw -> fresh wallet w'
w' executes trade for tau on Base