Asking good questions

Intelligence does not answer from a summary someone prepared earlier. When you ask it something it goes and queries your own records — the same trades, journal entries and emotion logs the Tracker reads — and then writes an answer from what came back. That round trip is why it can be checked, and it is also why the shape of your question matters more than its wording.

What it is allowed to look at

Intelligence has a fixed list of lookups and nothing else. There is no free-form query box behind it and no way for a question to widen its reach. Fifteen of those lookups read; the two that write are covered in their own article.

• Your headline figures — net P&L, win rate, profit factor and the rest of the summary, over the trades in scope.
• Breakdowns — performance grouped by symbol, direction, setup, weekday or hold time.
• Time of day — the weekday-by-two-hour-block grid of when your trading actually works.
• A day — the full daily report for one date, the same one Reports draws.
• Your trades — the list itself, filtered to winners, losers, one symbol, one direction or still-open positions.
• One trade in detail — including its risk figures, when it has named a specific trade.
• Your journal — which days you wrote up, and the text of a given entry.
• Emotions — the breakdown of how trades that felt a certain way performed.
• Strategies and distribution — your playbook and how each entry in it performed, and your winning, losing and breakeven counts by day, week or month.
• Your trading plan — the trades-per-period caps you set yourself, and how your actual weekday counts sat against them.
• The economic calendar — scheduled releases, for a question about a day the market was told something.
• What it remembers — the durable facts you have asked it to keep.

Each of those runs as you, against your own data, through the same code the equivalent page uses — so Intelligence can never read something you could not open yourself. A closed list is a small limit to accept for that guarantee.

Ask for the subset you actually mean

The card underneath an answer is the rows the lookup returned, not a selection the model made afterwards. So a question that names its subset up front — winners only, one symbol, shorts, still-open — gets a card containing exactly those. A vague question gets everything, and then a narrowing described only in the prose would disagree with the table sitting under it.

• Name the population: “my losing trades in NQ last month”, not “how did NQ go”.
• Name the axis you want it grouped by — symbol, direction, weekday, time of day, hold time, setup.
• Ask for one thing when you want one thing. A superlative without a limit still draws a whole table.
• Ask it to compare two named windows rather than asking whether you are “improving”.

How many lookups it gets

The Speed dial sets the budget: three lookups on Quick response, eight on Auto, twelve on Deep research. Once it is spent the tool list stops being offered at all — it is told to answer from what it already has and to say the answer may be partial, rather than looping. A question that needs three different angles is still better asked as three questions in the same thread, where each answer is grounded in its own read.

What a thread is reading

This is the one surface the navbar controls do not reach. A conversation is created with no account filter and no date window, so it grounds over every account you own and your whole history. That scope is stored as “everything” rather than as a list of accounts frozen on the day you started, which means an account you connect later is included in an old thread too. Ticking accounts in the picker or moving the date range changes the Tracker and the Portfolio; it does not change what a chat can see. Say “on my funded account” or “in July” in the question instead.

The two dials on the composer

• Speed — Auto, Quick response or Deep research. It reaches the server and picks both the model behind the turn and how many lookups that turn is allowed, so it genuinely changes the answer and not only how long you wait.
• Approach — Standard or Extended — whether it reasons in the open before answering. The control is there and holds your choice while the page is open, but it is not sent with the question, so it does not change the answer today.

What comes back

An answer in plain words, and under it the cards built from the lookups it ran — a summary, a bar chart, an equity curve, a table of trades, a list of releases. The model never writes those cards. It cites the lookups it made and the server renders the results it kept, which is what stops a card from showing a number the answer invented.

Answers also carry references back to what they were built on: a trade, a journal entry, a strategy, an account, a named metric or an economic release. Follow one early on. Checking two or three answers against the Trades table is how you work out what this thing is good at, and it takes a minute.