Most "whale alert" products stop at a notification. A wallet moved a large amount; you get a ping; then you're on your own to decide, execute, and — if you're disciplined — review. SolNexus Trade is built around a different assumption: a signal is only as good as the loop it lives in. This post walks through the full path a real on-chain event takes inside SolNexus — from raw wallet transfer to a scored, wallet-native trade and back into a model that learns from the outcome.
Detection: two engines, one goal
SolNexus runs two independent real-time engines. The TokenPricePoller pulls price and volume across thousands of tokens via Jupiter and DexScreener. The WalletAlertPoller watches whale and shark wallets directly on-chain through Helius RPC — pure transaction parsing, no third-party price APIs. Together they surface five signal types: price surges, price drops, volume spikes, smart buys, and smart sells.
Every event is captured with its transaction signature and block time. That signature is the anchor of everything downstream — the one piece of evidence you can independently confirm on Solscan.
Scoring: the ML pipeline (the part that matters)
This is where SolNexus diverges from alert-only tools. Before any signal is emitted, it passes through four scoring layers that each adjust a live confidence_score:
- L1 — a deterministic formula score from signal strength, liquidity, and volume ratios.
- L2 — historical reinforcement: if a signal type has at least 15 historical outcomes, the score is nudged up or down based on how well that type has actually performed (precision and median move).
- L3 — a contextual Thompson Sampling bandit. This is the core ML layer: Beta(α,β) posteriors are maintained per context bucket, so the system leans toward actions that have paid off and shrinks those that haven't.
- L4 — an execution-policy score that tunes how aggressively the bandit is applied.
Then an Emit Gate enforces three minimum thresholds — confidence ≥ 50, execution_confidence ≥ 58, token_quality ≥ 62. Signals that don't clear all three are silently dropped. You never see low-conviction noise.
Execution: wallet-native login (self-custody); live uses a dedicated Key Vault trade wallet
Passing signals reach your isolated bot worker, which polls every ~8 seconds and applies your personal config (strategy, direction, allowed signal types, confidence gates). When a position opens, execution is via Jupiter swaps. Paper needs no keys; for live you authorize a dedicated trade wallet whose key is secured in Azure Key Vault so the bot executes autonomously. Checkout is an on-chain USDC transaction you verify on Solscan. No card.
Review: the self-calibration loop
Most platforms end at the alert. SolNexus closes the loop. When a position closes, its realized P&L is recorded against the original alert and fed back into the ML pipeline. L2 reinforcement and the L3 bandit posteriors update automatically. The bot adapts to your market conditions with no manual retraining — a fresh bot starts conservative and earns its aggression as evidence builds.
On top of this, every signal is graded on +15m / +1h / +4h windows in the Review tab, so the misses are logged alongside the hits. That's an auditable system, not a highlight reel.
Why this design is rare
To our knowledge, no other Solana trading tool closes the entire loop — detect, ML-score, execute wallet-native, and self-calibrate from every closed trade — inside one workflow. Most bot platforms stop at the alert or at a static model. The learning loop is what makes SolNexus adaptive rather than brittle. (All of this is documented at solnexus.xyz/bot/guide — verify it yourself.)
Getting started
SolNexus is live today. The Free tier gives a live market overview and sample whale alerts on a 4-hour delay — enough to see the mechanism. Founding members can lock in $25/month for three months through the waitlist at https://solnexus.xyz/waitlist. Login is wallet-native via Phantom or another Solana wallet — self-custody by default.
If you want to see how a real on-chain whale flow becomes a verifiable, ML-scored signal, the desk is running now. Start at https://solnexus.xyz and claim founding access at https://solnexus.xyz/waitlist.
Top comments (0)