0xrhXBT

New chains launch faster than the tooling that makes them legible. Robinhood Chain is a fresh, fast-moving L2 with a genuinely novel asset mix — but no native intelligence layer to tell a human (or an AI) what is trending, what just launched, and what to distrust. 0xrhXBT is that layer. A pipeline ingests onchain and market data every few minutes, a transparent scoring engine ranks momentum, a launch radar surfaces new projects with rug heuristics, and an autonomous agent narrates it all to X. The design principle throughout is trust by construction: every published number must be derivable from a recorded source, and rugged or manipulated tokens are filtered before they can ever reach a reader.

Robinhood Chain (chain id 4663) is an Arbitrum-Orbit L2 where three worlds share one order book:

The signature phenomenon is a tokenized stock drifting away from its oracle price after hours — pure onchain price discovery that only exists here. That divergence is signal, and it is one of the first things 0xrhXBT reports.

One pipeline writes to a single embedded database; the terminal, the agent, and the API all read from it. No hidden services, no second source of truth.

The agent is the only writer; readers see fresh data within seconds over a Server-Sent Events stream.

Every metric is attributed to where it came from:

• ›public RPC — blocks, transactions, contract calls, token metadata
• ›Blockscout — holders, transfers, contract verification, address counts
• ›Chainlink — oracle prices for tokenized stocks — the reference for divergence
• ›DexScreener — community-token price / volume / mcap / liquidity from real DEX pairs
• ›DeFiLlama — chain TVL and new protocol listings
• ›twitterapi.io — launch chatter and KOL mentions for the radar (reads only)

Community tokens price from the DEX (what traders actually pay); canonical stocks price from the oracle. Keeping the two worlds honest against each other is the point.

The trending board is ranked by a single 0–100 trend score, not by raw price change. Each factor is a z-score against the token's own trailing baseline (so a $5m memecoin and an $800m stock are judged on their own terms), squashed through a logistic, then blended:

• ›wash-trading cap — 24h volume is clamped to 20× pooled liquidity before scoring, so a token can't fake momentum by round-tripping its own supply.
• ›EMA smoothing — scores are exponentially smoothed across hours, so one noisy print doesn't whipsaw the board.
• ›cold start — a token with <48h of history is ranked cross-sectionally against its peers and flagged warming up, never given false confidence.

The radar finds projects early — from X launch chatter (KOLs weighted heavier), our own onchain discovery of tokens under 48h old with small mcap, DeFiLlama listings, and freshly verified contracts. Every candidate is scored on a conservative safety heuristic built from objective onchain facts:

A high score means "nothing obviously wrong", never a guarantee — and the agent is required to phrase it exactly that way. A clean, ultra-early launch (or genuine KOL buzz) fires an instant alert.

A rugged token on the trending board would defeat the platform's purpose, so drained pools are filtered out before anyone sees them. The guard keys on the one thing a rug can't hide — the exit door closing:

• ✕a community pool drained to dust (<$2k liquidity), or
• ✕liquidity collapsed ≥85% from its 24h peak and now thin.

Canonical tokens (stocks / stables / majors) are exempt, unknown liquidity is never assumed rugged, and a price drop alone never hides a token — a good project can be down and stay listed. The same filter gates the agent's posts.

The agent's posts are drafted by a language model, but the model is never trusted with facts. Every generated post, alert, and reply is passed through a grounding validator: each number in the text must be derivable from the structured data payload the model was given (raw, or k/m/b abbreviated, within tolerance). If a number can't be traced to a source, the draft is rejected — it never reaches the queue, let alone X.

The agent runs on a fixed schedule; a human can review every draft in the terminal's queue before it publishes (and moderation/deletion lives behind an admin console).

Publishing is throttled (posts / alerts / replies each capped per hour) and replies are bounded per conversation, so the agent stays a signal, not a spammer.

Everything above rolls up to a few promises:

The terminal is fully usable for free today. Wallet login (Sign-In With Ethereum on Robinhood Chain) and paid tiers with deeper history, priority alerts, and higher API limits are coming soon.

not financial advice. 0xrhXBT reports data and momentum; what you do with it is your call.