Market Terminal

This is an independent build — an homage to the Bloomberg Terminal, not affiliated with or endorsed by Bloomberg L.P. I wanted to see how much of a dense, keyboard-first market workspace I could put together from free data sources: cross- asset quotes, charting with indicators, an asset-detail workspace, a virtual portfolio with real cost-basis maths, a macro dashboard and calendars, all behind a ⌘K command palette.

What's real, and what's simulated#

The first thing I designed was honesty about data, because a terminal that lies about its numbers is worse than no terminal. Every value carries a provenance tag — LIVE, SIM, or CACHE — next to the panel it came from. Crypto quotes and OHLC come from CoinGecko (no key needed, so crypto is live out of the box); stocks, ETFs, indices and FX from Twelve Data; quotes, profiles and news from Finnhub; macro series and the yield curve from FRED; headlines from NewsAPI. With no keys at all the app runs entirely on simulated data and every panel says so.

The simulator isn't random noise — it holds its invariants: bid below ask, the 52-week range brackets the last price, ownership sums to 100%, each macro series sits at its real-world level, no IPO prices on a Saturday. Plausible-looking nonsense is the failure mode that actually costs trust, so those invariants are pinned by tests.

Architecture#

The data flow is a single line, and I held to it strictly:

Component → hook (TanStack Query) → RoutingProvider → adapter → (on failure) → Mock

No component calls fetch() or the mock generator directly — everything goes through the data hooks, so swapping a source never touches a component. A missing or rate- limited provider falls back to the simulator automatically and the tag flips to SIM. Server state lives in TanStack Query; local truth (watchlists, the portfolio ledger, layout) lives in Zustand and persists to localStorage. API keys never reach the browser — adapters run server-side behind a single /api/data/[method] endpoint.

The portfolio#

The portfolio maths is real regardless of whether the instruments are live or simulated. Positions are replayed from a transaction ledger through FIFO lots, and every P&L figure derives from that. The tests are chosen to discriminate rather than merely pass — FIFO returns 1500 where average cost would say 1000. Beyond unit tests on the pure layers, every phase was driven end-to-end in headless Chrome, which is what caught the things types couldn't: a chart rendering with zero height, a container collapsed by its flex parent, a stale query surviving a reopened palette. It's dark-only by design, but the text ramp is measured against WCAG AA and charts never encode meaning in colour alone.