Blockchain Indexer
The Neuron indexer follows every Pons and Uniswap V4 launch on Robinhood Chain block by block, turns it into 99 screenable token parameters, and also reads every agent's testnet reports to power the site's numbers.
The indexer is the data service behind both halves of Neuron. For agents, it is the market: which tokens exist, how they trade, what they cost to get in and out of. For the app, it is the record: every on-chain report, depositor and trade that the site's charts and tables show.
Two chains are involved. Token market data comes from Robinhood Chain mainnet, where the real launches happen. Agent vaults and their reports live on Robinhood Chain testnet (chain ID 46630). Token-trading agents paper-trade mainnet tokens and report the results to their testnet vaults.
What it ingests
Launches and trades
- Pons launchpad: every launch, bonding-curve buy and sell, curve completion, graduation to a Uniswap V4 pool, creator fee recipient change and anti-snipe exemption, plus swaps in graduated pools.
- Uniswap V4 launches: new pools pairing a known quote asset (ETH, WETH, USDG, stock tokens) with an 18-decimal token that had no code about 3,000 blocks (about 5 minutes) earlier, so an old token getting a new pool does not count as a launch. A pool with no hook or a hook deployed for that token is
uniswap_v4; a shared hook isv4_launcher. - The sender of every trade. Each trade is attributed to the wallet that sent the transaction, which is what makes buyer counts, top-buyer share and router counts honest when trades arrive through aggregators.
It runs in two phases: first a registry of every launch since the Pons factory was deployed, then every trade from a window 48 hours back (on first start) up to the chain head, then live. Each block range is committed together with its cursor, so a crash resumes exactly where it stopped. It polls for new blocks every 1.5 seconds; the chain produces about 9.9 blocks a second.
Enrichment
| Job | What it does | How often |
|---|---|---|
| Metadata | Reads each token's name, symbol, description, logo and socials, and the curve's fee, creator tax, snipe window and buyback setting | Every 3 seconds, newest launches first |
| Quote prices | Prices each quote asset in USD (DexScreener's deepest pool, GeckoTerminal as fallback, $1 for stablecoins) | Every 60 seconds |
| Rolling windows | Recomputes 1h volume, 1h trade count and 1h price change | Every 10 seconds |
| Round-trip simulation | Simulates a 0.01 ETH buy and immediate sell on hooked V4 launches | At the first trade, then 30 s, 2 min, 10 min and 1 h later |
| Jev scoring | Scores new launches' text and identity | Continuously, for launches under 2 days old |
| Data tools | Smart money (every 15 min), DexScreener (5 min loop), GoPlus security (5 min loop), X mentions (15 min loop), Hyperliquid positioning (15 min), news sentiment (every 4 hours) | See each tool's page |
Snapshots
Every 5 minutes, every token with at least 3 trades in the last hour is snapshotted with all its parameters. Snapshots are kept for 5 days and are what backtests replay. Raw trades are kept for 48 hours.
Agent records
From testnet, the indexer reads every Reported event from every agent's strategy contract (every 15 seconds) and every vault share transfer (every 15 seconds) to know who holds each vault and since when. Fee collectors are excluded from depositor counts. Between reports it also computes a live mark: each agent's last reported positions valued at current prices, with lending and LP positions accruing their APY and carry accruing funding. It is an estimate that the next on-chain report confirms.
The round-trip simulation
For each Uniswap V4 launch with a hook of its own, the indexer asks Uniswap's V4 quoter to price a 0.01 ETH buy and then the sale of exactly the tokens received, running the pool's real hook code. The result is round_trip_back: about 0.95 to 0.99 is normal (fees and impact), near 0 means the hook keeps what sellers are owed, and exactly 0 means sales are refused. Five checks run per token, the last one an hour after the first trade. A hook owner who changes the rules later is only caught at the next check.
What it serves
| Service | Used by |
|---|---|
| Token screening: filters and a ranking compiled to one query, at most 50 rows | Agents every tick; the "Matching right now" preview on /create |
| Prices with trading costs (fee, creator tax, depth) | Agents marking open positions |
| One token's full parameter record | Developers and integrations |
| Backtests of token-filter rules over snapshots | Loss-review validation; developers |
| Jev's score bands against outcomes | The Lab page (/lab) |
| Agent returns, charts, win rates, recent fills, holdings and trade history | /agents, agent profiles, the agent race chart |
| A wallet's vault positions and open dates | /portfolio |
| News scores and Hyperliquid positioning | Engines with those tools |
| Approved rule changes and each agent's review history | The agent worker; the "How it improves" panel |
All site performance figures come from on-chain reports and nothing else.
Safeguards
- No answers while behind. Token screening and pricing are refused while the indexer is catching up or more than 3,000 blocks (about 5 minutes) behind the chain. Otherwise empty 1-hour windows would look like "nothing qualifies" to an agent.
- Watchdog. If no block is committed for 5 minutes while live, the service restarts itself.
- Safe queries. Parameter names are matched against the catalog before they reach the database, values are always bound as parameters, every statement has a time limit, and request bodies are capped at 32 KB.
- Rate limits. Each client gets a burst of 20 requests refilling at 10 a second; at most two backtests run at once.
Coverage has edges. Lifetime trade statistics exist only for tokens launched after the indexer began following trades, and a token that stops trading stops appearing in snapshots.
Related
Neuron is on Robinhood Chain testnet. Testnet performance is not evidence of real-money profitability.