Bidwiserai

Get the user's current holdings: quantities, cost basis, live value, unrealised gain/loss, position weights and sector breakdown. Optionally includes today's market events already scored for how much they actually affect this specific portfolio. Call this first for almost any question about the user's money — every other tool reasons against this snapshot.

Retrieve the reason the user originally recorded for owning a position, plus their stated goal and time horizon. Use this when a reflection needs to point back at the user's own words rather than at generic advice — it is the strongest material available for a cooling-off question.

Get what has actually happened in the market: index and sector moves, live quotes for the user's holdings, and the recent event feed. Use this to establish whether a move is market-wide, sector-level or specific to one company — that distinction is the single most useful fact when someone is reacting to a price drop.

Layer A of the guardian. Work out whether a specific market event actually touches this portfolio, by how much, and why — returning an exposure percentage, the affected holdings, and whether the move is broad-market, sector-level or company-specific. Use this before any intervention: a broad dip and genuinely bad company news deserve completely different conversations.

Quantify what the crowd is doing right now: the share of orders on each side over 24 hours, mention volume and how far above normal it is, plus a market-wide fear/greed reading. Use it when the user cites what other people are doing, so "everyone is selling" becomes a number they can look at. Crowd direction is descriptive only — it is never evidence that the crowd is right, and inverting it into a contrarian call is just as much a recommendation as following it.

Layer B of the guardian. Analyse the user's own words for one of three bias patterns: panic_sell, fomo_buy or herd_follow. Returns the pattern, a confidence score, the exact phrases that triggered it, and an action band (intervene / watch / none). Call this on any message where the user states an intent to trade — and trust a 'none' result: a calm, specific, pre-planned trade is not a bias event and must not be interrupted.

Pull what has historically happened after decisions like this one: aggregate outcomes for the cohort plus specific comparable episodes. Use it to give the user a base rate instead of an opinion. Includes deliberately-chosen counterweight cases where the reactive decision turned out to be the correct one — presenting only the cases that support pausing would be advice wearing a costume.

Compose the intervention: a plain-language observation about what is actually happening to this portfolio, a non-judgemental note on the pattern detected, the historical base rate, and exactly one reflective cooling-off question. This is the only tool that produces user-facing prose, and it lints its own output against a list of banned directive phrases before returning. Renders the intervention widget.

Look up this user's own past decisions: how many times each bias pattern has fired, how often they proceeded versus paused, and how those earlier calls turned out. Use it to make an intervention personal — "this is the 3rd panic-sell signal on this account this month" lands harder than any general statistic. Report the count as an observation, never as an accusation.

Show what happened to a mock cohort that waited 24 hours, 7 days or 30 days instead of acting immediately, for this type of market event. Returns the favourable AND unfavourable shares together — for company-specific news, waiting historically left the cohort worse off, and that must be shown too. Use it to give a base rate for the cost of a pause, never to argue that pausing is correct.

Record the full context that was shown to the user and the decision they made. Call this after every intervention, whatever the outcome — including when the user proceeds anyway. This is the audit trail that makes the intervention defensible, and the research data point that makes it improvable.