We don't publish testimonials. We publish the database.

Verified production telemetry from the live Loxley AI trading bot fleet on Robinhood Chain: 2,064 tokens safety-scanned, 1,557 unattended buys gated, 143 snipe bundles flagged, 989 confirmed on-chain trades across 458 tokens — with methodology.

Most trading bots market themselves with screenshots. This case study uses the opposite source: the production database of the live Loxley AI bot on Robinhood Chain, queried on 4 November 2026. Every figure below is a real count from the running system — the same engine operators get when they launch a white-label bot. The cohort is in its founding, micro-volume phase; we say that openly, and we publish nothing we cannot verify.

The safety engine is the headline

Between 20 September and 31 October 2026 the pipeline safety-scanned 2,064 tokens. Verdict distribution in the retained cache: 640 DANGER and 917 CRITICAL ratings. Unattended auto-buy was gated on 1,557 tokens — roughly three out of four scanned tokens were not fit for automation. On a launch-heavy chain, that ratio is not a bug in the market; it is the market. An engine without a gate would have put members into hundreds of those.

• 143 bundle / snipe-cluster detections — creations whose early supply sat with deployer-linked or same-block wallets (see the snipe detection documentation).
• 19.3 seconds average full scan time — that is the deliberate cost of deep verification: contract authorities, liquidity-lock proof read on-chain, hook inspection, simulation. Fast scanners skip layers; ours does not.
• Degraded verdicts are never cached: a scan that could not complete reports degraded and recomputes on next demand. Stale "probably fine" is the classic way bots buy rugs.

Execution telemetry

• 989 confirmed on-chain trades — 560 buys and 429 sells — across 458 unique tokens between 10 September and 4 November 2026.
• 429 confirmed exits executed under the standing rule that sells are never blocked by the safety scanner (sell engine docs).
• Position management ran unattended: trailing stop-losses and take-profit ladders evaluated server-side around the clock.

Infrastructure in production

The live fleet serving the cohort: 10 API worker instances + 1 coordinator + 12 trade executors (23 containers), backed by PostgreSQL and Redis — the sharded pusher/receiver model described in the architecture documentation. Alert fan-out, order execution and the safety pipeline run on isolated lanes so launch storms cannot starve exits.

The founding cohort, honestly framed

35 users are registered; 7 are actively trading. This is deliberately an early, micro-volume phase: the engine is being proven against live market chaos before the fleet is opened wider. We do not publish profit-and-loss claims for this phase — when the cohort operates at scale, realized-operator metrics will be added here with the same database-verified methodology. Anyone selling you a bot on profit screenshots during week three is selling you fiction; we would rather be believed.

Methodology

All figures are direct aggregates of the production database (trades table: status=confirmed; safety cache: verdict and gating counts; scan timings: recorded per scan). Snapshot date: 4 November 2026. No estimates, no projections, no cherry-picked winners. Nothing here is financial advice; trading is risky and past infrastructure performance guarantees nothing about future token outcomes.

What this means for operators

The white-label product is not a promise that members will win — it is an engine that (1) knows which tokens to keep automation away from, (2) never imprisons an exit, and (3) scales its messaging past Telegram's limits. Those are the properties an operator's brand survives on.

Run this engine under your brand → Read the technical docs

Built for operators, not slippage goblins. Talk to us →