OpenTerminalUI — Shipping 1.0, Where Integrity Is the Feature

After months of forking, rebuilding, and de-Indianizing a Bloomberg-style financial terminal, we tagged v1.0.0. The strange part — and the point of this post — is that 1.0 added almost no new features. We spent the entire release on a single, unglamorous promise: the terminal will never show you a fabricated number as if it were live.

This is installment 4 of the OpenTerminalUI series — the 1.0 milestone. The two deep-dive installments promised earlier (the wiring audit and the run-on-any-LLM story) are still on the way as #2 and #3; this one is the capstone.

Background: a terminal you're supposed to trust with money

OpenTerminalUI is a self-hosted financial terminal — live charts, screeners, portfolios, options analytics, an AI research layer — that you run on your own machine. Earlier installments covered why we forked it and rebuilt its India-centric data layer for US, European, and crypto markets.

The fork has a north star: an open, private terminal that helps an individual invest without being fooled — by the market, by hype, or by the tool itself. That last clause turns out to be the hard one. A weather app that guesses is mildly annoying. A financial tool that shows you a confident, precise, wrong number is dangerous, because precision reads as truth.

A couple of terms, defined once:

Mock / fallback data — placeholder values a program serves when the real source fails, so the screen isn't empty.
Degraded — our label for "we have no live data here right now," shown honestly instead of filled in.

The problem: software that lies politely

When we audited the backend before 1.0, we found a recurring, well-intentioned pattern: when a real data source was missing or rate-limited, the code made something up so the UI looked alive. The examples were sobering:

The commodities panel served 2100.0 + random() as a "live" gold price.
Charts had a synthetic random-walk fallback — backtests could run on a price history that never happened.
The order book was generated from a hash of the symbol name — a plausible-looking ladder of pure fiction.
The Relative Strength screens returned hardcoded Indian blue-chips, presented as live rankings.

None of this was malicious. It's the path of least resistance under "never show an empty panel." But for a tool whose entire reason to exist is not being fooled, polite lies are the worst possible bug. They don't crash, they don't log, they just quietly mislead.

What we did: make "I don't know" a first-class answer

The fix wasn't one patch; it was a convention and a sweep.

Bucket	What it covered
A — Integrity	Audit every fabrication site; wire real data or mark it degraded
B — De-India defaults	Western-market defaults so the release identity is coherent
C — Robustness	Back off from rate-limited APIs instead of hammering them
D — Version contract	One real version number, honestly surfaced
E — Docs	Write down, plainly, what works and what doesn't

The heart of it is bucket A. We established a single degraded-data convention: when there's no live source, the API returns an empty result plus a small machine-readable marker — a reason (no live source, missing API key, rate limited) and the source it tried. The front end reads that marker and shows a banner. The rule is one sentence: nothing fabricated may look live. Gold prices, charts, order books, relative strength — each fabrication site became either real data or an honest blank with a reason attached.

Around that, the supporting work:

Robustness (C). A shared retry-with-jitter plus a circuit breaker, so when a free-tier API starts returning "too many requests," the terminal backs off and serves cached data instead of pounding a provider that's already down. Responses from the data providers are now cached in a persistent layer, so identical requests don't re-spend a daily quota.
A real version contract (D). The project had drifted into two version numbers — the front end said 0.4.0, the backend said 0.2.0. We reconciled them to a single 1.0.0, made the front end derive its version from one source, and surfaced it in the health endpoint and the UI footer.
Honest docs (E). A "what works out of the box vs. what needs an API key" matrix, and a Limitations page that says out loud the uncomfortable things: there's no free live economic-calendar source (we ship a clearly labelled sample); dividend forward dates are estimates; several screens — bonds, movers, insider trades, the tape — are honest empty stubs until a real feed is wired.

Why integrity over features

It would have been more fun to add features. We deliberately didn't, for three reasons.

Trust is the product. For a research terminal, a feature that occasionally lies is worth less than zero — it poisons confidence in every other number on the screen. A blank panel that says "no live source" costs you nothing; a fake gold price can cost you a trade.
A version number is a promise. Calling something 1.0 says "you can build on this." We didn't want the first stable tag to sit on top of a layer of random-walk charts and hashed order books. 1.0 had to mean honest, not feature-complete.
Empty is recoverable; fake is not. A degraded marker is a to-do list — it tells you (and future contributors) exactly where a real data source still needs wiring. Fabricated data hides that gap and makes it look solved.

Impact

The terminal now fails honestly. Where it has live data — US/EU/crypto charts and quotes via free sources, crypto fundamentals, the local-LLM research layer — it shows it. Where it doesn't, it says so, with a reason, instead of inventing a number. For someone using it to make real decisions, that's the difference between a tool and a toy. And for the next contributor, the degraded markers are a precise map of what's left to build.

Scope: what 1.0 is not

Honesty about the edges, in the spirit of the release itself:

The degraded stubs are still stubs. Relative Strength, bonds, hotlists, insider trades, and the time-&-sales tape return honest empty results — the real data engines behind them are future work, not part of 1.0.
No free Level-2 depth for US/EU equities exists, so that order book is empty-and-labelled rather than faked; a brokerage-API adapter is planned.
The release smoke-test matrix is manual for now. We verified the core flows by hand across database and key combinations; automating that matrix is a deliberate to-do, not a silent gap.

What comes next

v1.1 — multi-asset depth: widen the real coverage (EU/crypto heatmaps, portfolio cash and transactions, more live order-book sources).
v1.2 — AI-native deepening: lean into the private "second brain" research layer that grounds answers only in your own notes.
And the two series deep-dives still owed from installment 1: the wiring audit ("the bug that looked like 33 bugs") and the run-on-any-LLM rebuild.