Quiet software, loud money.
Three years of building trading systems and data pipelines on Solana and Ethereum, compressed into the lessons we keep re-learning.
For three years a good share of our client work has lived on Solana and Ethereum: live trading tools, indexers, pipelines that turn raw chain data into decisions. On-chain, a bug isn't a ticket in a backlog — it's a number going down in someone's wallet. Here's what that pressure taught us, in the order it hurt.
Latency is a feature
When markets move in milliseconds, being right slowly is the same as being wrong. We learned to treat latency budgets the way other teams treat feature specs: measured per hop, owned by someone, regressed in CI. The fastest code is the code that isn't there — half our optimization wins were deletions.
The corollary: you can't bolt speed on later. Every system we tried to "make fast afterwards" got rebuilt. Every system designed around its hot path from day one is still running.
Test like the chain is hostile
RPC nodes lie. Blocks reorg. Transactions vanish from the mempool and reappear two blocks later with different neighbours. The happy path is a rounding error of what production actually serves you. So we simulate the ugly parts — dropped connections, stale reads, duplicate events — and make the system prove it survives them before it ever touches a real key.
The chain doesn't care about your happy path. It only ships the unhappy ones.
Boring wins
The systems still running years later aren't the clever ones. They're the boring ones: idempotent jobs, explicit state machines, alerts that fire before the client notices, dashboards a stranger can read at 3am. Cleverness is a liability you pay interest on every time you're paged.
That's the whole philosophy, really — quiet software, moving loud money. If you're building on-chain and it feels calm, either you've done it right or you haven't shipped yet.