How Loxley AI detects snipers on fresh Robinhood Chain launches: creation-transaction anchoring, deployer and LP-funder linkage, same-block buy analysis, concentration scoring, and how verdicts feed the safety gate.
Every new token on Robinhood Chain leaves a forensic trail in its first blocks. Loxley AI's snipe detection module reads that trail to answer the question that decides whether a launch is tradeable: is the early supply in the hands of a fair crowd, or of the deployer's friends?
The analyzer anchors every token to the transaction that created it (or to its first liquidity event, when the launchpad flow differs). From that anchor it reconstructs the launch graph: the deployer address, the wallets that funded the deployer, the liquidity provider and the LP funders, and every buy executed in the same block and the blocks immediately after. Anchoring to creation is deliberate: later events can be padded with fake "community" buys, but the creation window is the one record the scammer cannot rewrite retroactively.
The module produces a concentration score and a set of labels (deployer, insider, LP-linked, bot-cluster) that flow into the layered safety verdict described in the scanner documentation. A token can have impeccable contract code and still rate as dangerous because the supply distribution says the trade is already over before it begins.
Launch ecosystems have repeat offenders. Addresses and funding clusters associated with previous rugs are tracked, and new creations linked to them are flagged at detection time — before members spend a cent. For community operators this is the difference between a chat full of winners and a chat full of exit liquidity: the warning travels with the alert.
Deep linkage analysis is expensive, so the pipeline is ordered: cheap deterministic checks (locks, authorities, same-block concentration) run inline at detection speed; the heavier graph analysis runs asynchronously and upgrades a verdict within seconds. If the deep layer is incomplete when a decision is forced, the verdict reports exactly which layers are pending — degraded states are never hidden and never cached.
Next: sharded architecture → All docs
Built for operators, not slippage goblins. Talk to us →