Risk notice

Read this first. Velox is a research tool, not financial advice. Every claim below applies to you, regardless of what the signals look like or how good the track record appears.

Perpetual futures are high-risk

Leveraged perpetual futures carry the risk of total capital loss — not just your margin, but in extreme cases more than you deposited depending on the exchange's liquidation mechanics. A single bad trade with insufficient stop-loss discipline can wipe an account.

Velox's risk sizing caps per-trade exposure at 40% of portfolio using a Kelly-based formula, but:

Kelly assumes stable win rates. Markets change regimes; Kelly can over-size during a hot streak that's about to end.
Stop-losses are not guaranteed. During flash crashes or exchange outages, stop-loss orders may fill well below (or above, for shorts) the trigger price. Slippage is your loss.
Funding rate risk. Perpetuals accrue funding payments; a position held against extreme funding can bleed P&L even while the price thesis is correct.

This is not financial advice

Nothing in Velox constitutes investment advice, a recommendation, or an offer to buy or sell any financial instrument.

Any signals, consensus tiers, or AI-generated reasoning are outputs of a computational process. They reflect correlations in historical data as interpreted by large language models. They are not predictions, and past performance of the engine (demo or live) does not predict future results.

Do not trade real capital based on Velox output unless:

You fully understand leveraged perpetual futures
You can afford to lose the entire capital you commit
You have independently verified the signal against your own analysis
You are aware of your local jurisdiction's regulatory stance on crypto derivatives

Model risk

Velox uses three large language models (Claude, Gemini, Grok) for analysis. LLMs have known failure modes:

Hallucination — models may cite indicators or price levels that don't exist in the input data
Training-data staleness — models don't know about market events after their training cutoff
Correlated failure — all three models trained on overlapping public internet data; they are not fully independent signals
Prompt injection — if input data contains adversarial strings (unlikely but possible in social sentiment pipelines), model output becomes untrusted

Consensus across three models reduces but does not eliminate these risks.

Operational risk

The pipeline runs on Vercel + Upstash Redis. Dependencies include:

Vercel (hosting, crons, serverless runtime)
Upstash Redis (shared state)
Anthropic, Google, xAI (model APIs)
Binance / exchange APIs (market data)

A failure in any of these can cause missed scans, stale data, or stuck trades. Self-hosted deployments should add alerting for:

Cron invocation failure (Vercel dashboard)
Redis unreachability
Model API timeouts or rate-limits
Market-data feed gaps

No warranty

Velox is provided "as is", without warranty of any kind, express or implied, including but not limited to warranties of merchantability, fitness for a particular purpose, or non-infringement. In no event shall the authors be liable for any claim, damages, or other liability arising from the use of Velox.

If you only read one paragraph

Treat any Velox output as a hypothesis you need to test against your own understanding of the market. The tool exists to surface ideas from a different angle than pure-discretionary trading — not to replace your judgment. If you wouldn't take the trade based on your own analysis, the fact that three language models agreed on it should not be sufficient.