What We Learned Building a Paper Trading Platform

Desk notes from client work. Verify claims independently — not a verified account of any production system.

Building a paper-trading product looks glamorous from the outside — charts, ticks, green and red. The hard parts are almost never the charts.

Market data is a contract, not a feed

Every vendor has a different definition of “realtime,” delayed, and corrected. We treat the feed as a contract: symbol universe, session calendar, corporate actions, and what happens when a tick arrives out of order. The UI only renders what the contract guarantees.

If you skip this, you will invent bugs that look like “the chart jumped” and waste a week blaming Recharts.

Ledger correctness beats flashy fills

A paper account is still an accounting system. We use double-entry from day one: cash, positions, and a fills journal that reconciles. Fuzz tests randomly buy/sell/split/dividend and assert the books balance.

Pretty order tickets without a correct ledger will lie to users — and to you — about P&L.

Realtime fan-out is a budget

We keep a 100ms budget from exchange tick to phone screen and account for every hop: ingest, normalize, shard, fan-out, render. Most “latency” complaints are actually reconnect storms or main-thread work in the client.

Boring is the product

KYC, statements, tax lots, session calendars — these are the features that keep a trading app alive past early access. We schedule them as fixed-price sprints with a crisp definition of done, not as “phase two.”

---

Want the short version? Rehearse the product on Stock Picks — it’s the platform we learned this on.