A typical piece of retail research goes like this. A question occurs to you. You search the web and get results from three years ago. You open a charting site to check the price. You open a screener to find comparables. You open a news site. Then you open your broker app, by which point you are placing an order on a reduced version of a thesis you can no longer fully reconstruct.
The cost is not the switching. It is that context dies at every boundary. Each tool starts from nothing, and the reasoning that justified the decision has evaporated by the time you make it.
Here is what a single continuous thread looks like instead.
Start with the question, not the ticker
Most research begins as a question in plain language, not as a symbol. "Why is this sector down today." "What are the expiries available on this index." "Which large-caps moved on volume."
The PocketX assistant takes the question in that form and answers from current PocketX market data — the same instrument, quote, screener, option chain and news services that power the rest of the product. It is an interaction layer over real data, not a general-purpose chatbot guessing from training data.
Two properties matter more than the answer itself.
The evidence is inspectable. Replies carry instrument references that turn into navigation. You can go from a claim to the instrument behind it and check.
Absence is stated. Where data is stale, the market is closed, or something is unavailable, the answer should say so rather than filling the gap. Backend timestamps are preserved. An answer that admits a limit is more useful than one that hides it.
Move into the evidence
From an answer, the thread continues into the surfaces rather than restarting.
- A stock reference goes to the instrument page — quote, price history, fundamentals, related news and contracts.
- An index reference goes to indices with its constituents.
- An options question goes to the option chain with available expiries and contract detail.
- A market-wide question goes to the heatmap or the screener.
This is the part that fragmented workflows lose. You are not re-typing a symbol into a different tool with different conventions; the instrument identity carries through, and so does what you were trying to establish.
Keep deterministic lookup one action away
An important design point that gets missed: the assistant is not the only entry.
When you already know exactly what you want, search remains directly available. Asking a question to reach a stock you could have typed the name of is slower, not faster.
Natural language is for questions you cannot express as a lookup. Lookup is for when you can. Both stay available, and choosing correctly between them is most of using the product well.
Where research becomes a rule
If what you found is a repeatable condition rather than a one-off observation, it belongs in the strategy workspace instead of your memory.
The path is deliberate: write the rule, validate it, backtest it, read the data assumptions, and attach the proof to the saved strategy. Evidence comes before activation — that ordering is enforced, not suggested.
Turning a hunch into a testable rule covers getting from an observation to something that can be tested, and seven ways a backtest lies to you covers reading the result honestly.
The boundaries that do not move
This is where the thread stops being seamless on purpose, and the discontinuity is the point.
The assistant is read-only. It explains market data. It does not place orders, and a generated answer is not individualised investment advice. Explanation and advice are different things, and the product keeps them separate.
Arming a cash strategy creates alerts only. It evaluates closed bars and notifies. It does not trade.
Strategy alerts place no broker orders. Nothing reaches a broker.
Live strategy order execution is excluded from the first release. There is no configuration that makes the strategy workspace place a real trade.
So every actual order is placed by you, through the order ticket, passing route and margin review. Contextual Buy and Sell actions across the surfaces reuse one instrument-seeded ticket, so the instrument carries through — but the decision does not become automatic at any point.
That gap between research and execution is not friction anyone forgot to remove. It is the product declining to blur the line between understanding something and acting on it. Smart alerts and the execution boundary sets it out in full.
Portfolio context, honestly
For active clients, a portfolio layer opens from any research surface without losing the route — broker orders, trades, positions, holdings and funds.
The honesty rules there are worth stating because they are unusual:
- Current valuation and P&L appear only when the broker inputs are complete. Partial data is not presented as a total.
- Historical performance stays explicitly unavailable until snapshot and cash-flow coverage exists. It is not estimated.
- Missing financial values are shown as unavailable, never as zero.
A missing number displayed as zero is worse than a gap, because it looks like information. Refusing to coerce it costs a cleaner-looking screen and buys a screen you can trust.
The thread, end to end
- Ask the question in the form it occurred to you.
- Inspect the evidence behind the answer.
- Widen or narrow — heatmap for structure, screener for names, chain for contracts.
- Check the news for what the data cannot say.
- Watchlist what is interesting but not actionable today.
- Formalise a repeatable observation as a rule, with proof attached.
- Decide and place the order yourself, with the position sized deliberately.
Steps one to six preserve context. Step seven deliberately does not automate.
Why continuity is the actual feature
None of the individual surfaces is unusual. Screeners, chains, heatmaps and news exist everywhere.
What is unusual is that the instrument identity and the research thread survive between them — across routes, reloads and devices, through real server-side sessions rather than a chat history that dies with the tab.
The practical result is that you arrive at the decision with the reasoning intact. Most bad retail trades are not analytical failures. They are decisions made on a fragment of a thesis, because the rest of it was lost somewhere between the fourth tab and the fifth.