Execution Intelligence

This document covers the four upgraded execution subsystems added to Sentry 2.0: the multi-model tip engine, the Orca Whirlpool swap integration, the dual-rail simultaneous submission system, and the RPC cluster consistency checker.

1. Multi-Model Tip Engine

Motivation

A single AI provider creates a single point of failure. If Groq is overloaded during a high-congestion event, every execution decision degrades to a static heuristic at exactly the wrong moment.

Sentry 2.0 now implements a provider cascade:

Groq (llama-3.3-70b-versatile)
  -> Anthropic (claude-3-5-haiku-20241022)
  -> Google Gemini (gemini-1.5-flash)
  -> OpenAI (gpt-4o-mini)
  -> Deterministic heuristic fallback

The first provider to return a valid JSON tip recommendation wins. If all configured providers fail, the heuristic fallback computes a tip deterministically from the tip floor and failure rate with no external call.

Hard Bounds

The following bounds are enforced regardless of what the AI returns:

Bound	Value	Rationale
Floor	1,000 lamports	Zero-tip transactions always fail under load
Normal ceiling	5,000,000 lamports	Prevents overspending during model hallucination
High-failure ceiling	`tip_max` from Jito percentiles	When failure rate exceeds 50%, allows full tip escalation

Network State Inputs

interface NetworkState {
  currentSlot: number;
  tipMin: number;
  tipMedian: number;
  tipP75: number;
  tipMax: number;
  recentFailureRate: number;       // 0-100
  timeSinceLastSuccessSecs: number;
}

API

POST /api/tip-decision
Content-Type: application/json

{
  "currentSlot": 320000000,
  "tipMin": 1000,
  "tipMedian": 5000,
  "tipP75": 15000,
  "tipMax": 100000,
  "recentFailureRate": 12.5,
  "timeSinceLastSuccessSecs": 30
}

Response:

{
  "recommendedLamports": 9500,
  "reasoning": "moderate failure rate, scale above median",
  "confidence": "high",
  "provider": "groq/llama-3.3-70b-versatile"
}

Implementation

lib/tip-engine.ts — provider cascade + classifyTxFailure() error taxonomy.

2. Orca Whirlpool Swap

Motivation

The primary demonstration of Sentry acting on Solana data rather than merely observing it. Using the official @orca-so/whirlpools-sdk, Sentry can execute real concentrated liquidity swaps against Orca Whirlpool pools on mainnet, with the resulting transaction submitted via Solami Beam.

Supported Pools

Pair	Pool Address
SOL/USDC	`HJPjoWUrhoZzkNfRpHuieeFk9WcZWjwy6PBjZ81ngndJ`
SOL/USDT	`4fuUiYxTQ6QCrdSq9ouBYcTM7bqSwYTSyLueGZLTy4T4`
SOL/mSOL	`9vqYJjDUFecLL2xPUC4Rc7hyCtZ6iJ4mDiVZX7aFXoAe`
mSOL/USDC	`AiMZS5U3JMvpdvsr1KeaMiS354Z1DeSg5XjA4yYRxtFf`

Execution Flow

WhirlpoolContext.from(connection, wallet, programId, configId) initialises the SDK with the Solami private RPC connection.
client.getPool(poolAddress) fetches live pool state.
swapQuoteByInputToken(...) computes the swap quote with slippage tolerance expressed as { numerator, denominator }.
whirlpool.swap(quote).build() produces an unsigned transaction.
The transaction is signed and submitted to Solami Beam via sendTransaction JSON-RPC.

API

POST /api/orca-swap
Content-Type: application/json

{
  "poolAddress": "HJPjoWUrhoZzkNfRpHuieeFk9WcZWjwy6PBjZ81ngndJ",
  "aToB": true,
  "amountLamports": 1000000,
  "slippagePct": 0.5
}

Response:

{
  "signature": "5xPq...",
  "inAmount": 1000000,
  "estimatedOut": 184320,
  "pool": "HJPjoWUrhoZzkNfRpHuieeFk9WcZWjwy6PBjZ81ngndJ",
  "route": "orca_whirlpool"
}

Implementation

lib/orca-swap.ts — swap execution and getWhirlpoolPrice() price oracle.

3. Dual-Rail Simultaneous Submission

Motivation

Sequential fallback (try Beam, then Jito, then RPC) adds latency at exactly the worst moment. During a congestion spike, the first provider may time out after several seconds, at which point the blockhash has aged and the retry lands in a degraded state.

The dual-rail pattern broadcasts the exact same pre-signed transaction bytes (identical 64-byte Ed25519 signature) simultaneously across Solami Beam SWQoS priority TPU and secondary RPC rails using Promise.allSettled.

Because identical transaction bytes share the same signature, Solana validators enforce protocol-level signature deduplication in their 150-slot recent blockhash cache. Whichever rail reaches the leader first commits the transaction; subsequent arrivals are dropped by the validator without double-spending. Each route's submission acknowledgement is tracked independently, and cancelling secondary HTTP sockets cleans up local network resources without affecting on-chain finality.

Submission Order

Rail	Endpoint	Method
Beam	`BEAM_ENDPOINT`	`sendTransaction` JSON-RPC
Jito	`JITO_BLOCK_ENGINE_URL/api/v1/transactions`	Bundle HTTP API
RPC	`SOLANA_RPC_URL`	`sendTransaction` JSON-RPC

Usage

import { submitDualRail } from "./lib/dual-rail";

const result = await submitDualRail({ serializedTxBase64: tx.serialize().toString("base64") });
// result.winner === "beam" | "jito" | "rpc"
// result.landingMs — milliseconds to first confirmation

Implementation

lib/dual-rail.ts — submitDualRail() and signAndSubmitDualRail().

4. RPC Cluster Consistency Checker

Motivation

A concrete, measurable demonstration that Solami private RPC consistently observes slots ahead of public endpoints.

What is Measured

Each probe fires three parallel JSON-RPC calls per node (getSlot, getVersion, getLatestBlockhash) and records:

Slot head at processed commitment
Response latency (ms)
Solana core version
First 8 characters of the latest blockhash

Interpretation

The solamiAdvantageSlots field in the report is the difference between the Solami slot and the arithmetic mean of all public endpoints. A consistent positive value confirms that the Solami private RPC is observing slots earlier than public infrastructure.

API

GET /api/rpc-check

Response:

{
  "timestamp": 1728460000000,
  "fastestNode": "Solami (Private)",
  "headSlot": 320485012,
  "solamiAdvantageSlots": 3,
  "results": [
    { "name": "Solami (Private)", "slot": 320485012, "slotLag": 0, "latencyMs": 28 },
    { "name": "Mainnet-Beta (Public)", "slot": 320485009, "slotLag": 3, "latencyMs": 210 }
  ]
}

Implementation

lib/rpc-consistency.ts — checkRpcConsistency() and formatConsistencyReport().

5. Dashboard UI Integration

All four execution intelligence modules are surfaced in the Sentry 2.0 Observatory Dashboard under the Intelligence and Execution Tools section (#intelligence-section), accessible directly from the top navigation dropdown and mobile menu.

5.1 Interactive Tip Decision Console (`TipDecisionPanel`)

Binds directly to live Solami private RPC telemetry (slot progression and 75th-percentile tip rates).
Allows operators to adjust simulated network variables (failure rate %, time elapsed since last confirmed transaction, minimum tip, maximum tip).
Displays the winning AI model from the provider cascade (Groq LPU, Anthropic, Gemini, OpenAI, or deterministic fallback), recommendation in lamports, reasoning explanation, and confidence classification (HIGH, MEDIUM, LOW).
Maintains an append-only in-memory log of recent execution decisions.

5.2 Real-Time RPC Consistency Benchmark (`RpcConsistencyPanel`)

Probes Solami Private RPC against public Mainnet-Beta and secondary RPC mirrors on demand.
Surfaces an empirical Solami Slot Advantage banner highlighting the positive lead delta (+N slots ahead of public cluster averages).
Tabular breakdown comparing slot head, lag relative to fastest node, ping round-trip latency in milliseconds, and solana-core software versions.

5.3 Orca Whirlpool DEX Execution Terminal (`OrcaSwapPanel`)

Enables manual and automated concentrated liquidity trades directly from the web interface.
Configurable liquidity pool selector (SOL/USDC, SOL/USDT, SOL/mSOL), trade direction (Sell Token A / Sell Token B), size in SOL, and slippage tolerance.
Dispatches signed swap instructions through Solami Beam SWQoS with live transaction status updates and one-click Solscan block explorer verification links.

5.4 Dual-Rail Live Submission Stream (`DualRailPanel`)

Connects to the real-time Server-Sent Events stream (/api/events) to monitor transactions submitted across the network.
Inspects winning execution rails (BEAM, JITO, or RPC) as Promise.allSettled races resolve.
Renders landing durations (ms), truncated transaction signatures, and confirmation timestamps in real time.

1. Multi-Model Tip Engine​

Motivation​

Hard Bounds​

Network State Inputs​

API​

Implementation​

2. Orca Whirlpool Swap​

Motivation​

Supported Pools​

Execution Flow​

API​

Implementation​

3. Dual-Rail Simultaneous Submission​

Motivation​

Submission Order​

Usage​

Implementation​

4. RPC Cluster Consistency Checker​

Motivation​

What is Measured​

Interpretation​

API​

Implementation​

5. Dashboard UI Integration​

5.1 Interactive Tip Decision Console (TipDecisionPanel)​

5.2 Real-Time RPC Consistency Benchmark (RpcConsistencyPanel)​

5.3 Orca Whirlpool DEX Execution Terminal (OrcaSwapPanel)​

5.4 Dual-Rail Live Submission Stream (DualRailPanel)​

1. Multi-Model Tip Engine

Motivation

Hard Bounds

Network State Inputs

API

Implementation

2. Orca Whirlpool Swap

Motivation

Supported Pools

Execution Flow

API

Implementation

3. Dual-Rail Simultaneous Submission

Motivation

Submission Order

Usage

Implementation

4. RPC Cluster Consistency Checker

Motivation

What is Measured

Interpretation

API

Implementation

5. Dashboard UI Integration

5.1 Interactive Tip Decision Console (`TipDecisionPanel`)

5.2 Real-Time RPC Consistency Benchmark (`RpcConsistencyPanel`)

5.3 Orca Whirlpool DEX Execution Terminal (`OrcaSwapPanel`)

5.4 Dual-Rail Live Submission Stream (`DualRailPanel`)