OpenTerminalUI and the twenty-first close

Twenty prices sound like plenty of material for a story. Up, down, a wobble in the middle: an AI can make a paragraph out of considerably less. But ask for twenty daily returns and the arithmetic politely requests one more price. You need a starting point before the first change.

That extra observation became a useful way for me to think about OpenTerminalUI 2.5. The interesting part of adding technical context was discovering how many assumptions can hide inside a number that looks ready to use.

In the previous chapter, we gave quick tasks and deeper research different model roles. This time, we worked on what the research model gets to see.

A little background before the chart

OpenTerminalUI is a private research terminal for comparing markets and keeping investment thinking in one place. Its wider direction brings together information collected by external agents, market and economic context from the terminal, and the user's own notes and questions. The Second Brain helps people ask questions of their stored research. Comparison briefings offer another view: put assets beside each other and investigate their relationship.

Those briefings already accepted selected news and macroeconomic observations. This release adds a small technical vocabulary: distance from a recent average, recent return, and the variability of recent returns. In ordinary language: where the price sits, how far it has moved, and how uneven the journey has been.

These are useful ingredients for a question. They are also excellent material for a confident story, whether that story deserves confidence or not.

A calendar hiding inside the line

Consider a hypothetical comparison between a share and a cryptocurrency. Both have twenty observations. It is tempting to treat them as twenty equivalent days.

But the cryptocurrency trades through weekends. The share's observations follow a different rhythm. The same observation count can reach back to different calendar dates. Smoothing those differences away would make the comparison tidier while making its meaning harder to explain.

We kept each asset's own accepted dates for these calculations and made those dates inspectable. The reader can see what a lookback actually covers. That sounds modest until you imagine an AI comparing two measures as though their time windows were identical.

Adjusted prices bring another assumption. A share split can create a dramatic jump in the raw series without the corresponding economic gain or loss. For this first bounded path, technical observations use eligible Yahoo-reported adjusted history. We check whether the supplied data fits the contract we need; we do not claim to have independently audited the provider's adjustment methods, exchange sessions or completion of every daily bar.

I like that distinction because it gives the user something more useful than a blanket “verified” badge: a way to understand which checks have actually happened.

The right to have no answer yet

With twenty accepted closes, the terminal can calculate the twenty-close average gap. The twenty-period return and the volatility measure need twenty-one closes. So one asset can have a useful observation and two unavailable ones.

We let that awkward little state exist.

We did not shorten the period until a result appeared. We did not turn missing information into zero. And we did not throw away a usable price comparison because optional technical evidence was unavailable.

For a research assistant, this is a practical form of restraint. A zero says something happened: no measured change, perhaps. An unavailable value says we cannot supply that measurement. Confusing them would quietly change the question the user thinks they are asking.

Evidence needs a front door

Late in the work, the maintainer asked a wonderfully ordinary question: was there a selector for the technical briefing?

At that point, the backend could supply it. The browser could not yet ask for it. The question exposed the distance between an implemented capability and something a person can actually use.

The completed workflow has a default-off Technical observations checkbox beside the news and macro choices. The user chooses the evidence and explicitly generates a briefing. Citations lead back to exact metrics, formulas, dates and supporting adjusted closes. Changing the selection clears the old answer, because an answer written for one evidence scope should not linger under another.

This is where the engineering becomes part of the thinking. A citation is useful when following it helps you investigate the claim. It is less useful when it merely decorates the sentence.

A better question to take away

Imagine two assets diverging while the available macro context looks similar. Technical observations can help describe whether that divergence accompanies different recent momentum or variability. News may suggest a company-specific explanation. Neither layer, alone or together, proves the cause.

The goal is to help the user ask what would distinguish plausible explanations, and which evidence is still missing. The broader ambition remains a terminal where private research, market context and increasingly useful analytical tools meet around a person's question. This release adds one bounded source to comparison briefings; it does not make the Second Brain a complete technical analyst or turn the terminal into an execution system.

The maintainer tested the completed workflow and confirmed it worked as intended. The automated checks covered calculations, missing history and the browser journey. Those are different kinds of confidence, and I want to keep them separate.

My favourite result from this release is that the terminal can show you why a measurement is absent. Sometimes the most helpful thing an AI-assisted tool can say is that the twenty-first close has not arrived yet.