The MarketXED in-app copilot is a constrained assistant for the product in front of you. It can explain committee mechanics, session state, and how a card is built. It must not invent a live ticker for a user who has not paid, and it must not pretend to be your broker. Use it as a glossary with context, not as an order router. Not financial advice.
What it is allowed to do
When you are subscribed, it may talk about the signal context the app already loaded (the same prints you can see on the tape). When you are not, it should discuss structure only: how agents vote, what a time stop is, when the cash session is open, how a 24-hour pass works. Empty fields mean paywalled, not “the market is empty.”
- Ask “how is conviction calculated?” not “what do I buy?”
- Ask “why would a timer exit be labeled HIT?” and then read the calibration piece.
- Ask “what does fractured agreement mean?” and keep the committee article open.
Strict answers stay short and finished. The model is instructed not to open a skeleton outline it cannot complete. If a reply truncates mid-claim, ignore it and ask one narrower question.
What it must refuse
Live entries, stops, targets, and “top ticker” lists are the paid object. A clever prompt does not override that. The copilot should not ask for passwords, OTP codes, or card numbers. It should not claim it placed a trade. If something in chat conflicts with the playbook card, the card wins until you understand why they differ.
How it fits the rest of the app
The copilot does not replace Wyckoff, VADER, or SMS. Those have their own pages so we do not recook the whole product in every paragraph. Use chat to unblock a definition, then go back to the chart. If you need a full day of live context, that is the 24-hour pass, not a longer prompt.
Feedback can sharpen wording over time, but it does not waive PDT, cash settlement, or the regular-hours SMS window. Those are policy and broker math, not chat style.
A good first session
Open the WhitePaper section you care about, ask the copilot to restate that mechanism in one paragraph, then compare it to a live card (if unlocked). If the two stories diverge, believe the card and file the mismatch. That is how you use an assistant without outsourcing judgment. When you are done, close the drawer; a copilot left open becomes another feed to narrate the tape.
Prompt patterns that waste the budget
Do not paste a screenshot novel and ask for “a full plan with sections.” Strict mode has a token budget and will refuse a skeleton. Ask one mechanism, get one finished paragraph, stop. Do not jailbreak for tickers. Do not ask it to override a stop on the card. If you need more depth, open the matching insight and quote a sentence you disagree with.
If chat and the WhitePaper collide, file it as a product bug rather than averaging them. Assistants drift. The playbook math and the committee meters are the system of record for a session. The copilot is a narrator with a paywall. Treat it that way and it stays useful.
Worked example: free user asks for today’s ticker
The copilot should refuse the name and point at how the committee works, plus the 24-hour pass if you want the tape. If it recites a ticker anyway, treat that as a leak, not a gift. Ask it to explain fractured agreement in one paragraph, then close the drawer and look at the meters yourself.
Keep a swipe file of good copilot answers that match the WhitePaper. When an answer drifts, you have a regression test. That is more useful than asking the same “what is conviction” prompt until the wording feels warmer. Warmth is not a mechanism. The meters are.