Smart money with Datadash

Use the Datadash MCP tools. Call get_schema once before composing any query, and keep the result; do not call it again in the same task.

Who counts as smart money

Unless the user gives their own criteria, smart money is a top-ranked wallet set over the whole history (duration ALL), applied as a lookup filter on the wallet field:

• For one event or market — the top 1000 wallets in the event's two main categories:

  {"field": "userId", "operator": "userTagProfileLookup", "value": {
    "tagId": {"operator": "in", "value": [<tag1>, <tag2>]},
    "duration": {"operator": "in", "value": ["ALL"]},
    "filter": [{"field": "rank", "operator": "lte", "value": 1000}]}}

  To choose the tags, find the event (a market's row carries its eventId) and read its tagIds; then rank those tags with query_lookup on a tagIds field, filtering id to the event's tagIds and ordering by numEvents descending, and keep the first two.

• Across all markets — the top 1000 wallets overall:

  {"field": "userId", "operator": "userProfileLookup", "value": {
    "duration": {"operator": "in", "value": ["ALL"]},
    "filter": [{"field": "rank", "operator": "lte", "value": 1000}]}}

The filter goes in walletFilters on unifiedSmartMoneySummary, unifiedSmartMoneyHolders and globalSmartMoney, and in filter on signalScore. A saved cohort can be used instead with {"field": "userId", "operator": "inCohort", "value": "<cohortId>"}. Always state which wallet set you used (the categories, duration and rank cutoff, or the cohort's name).

Find the event or market

Names are not ids. Resolve an event or market the user names with query_lookup (the event or market lookup table through an eventId or marketId field, with request.searchKey), and if several results match, show them and ask. For a Polymarket URL, search on its slug.

How smart money is positioned on an event

1. Summary — unifiedSmartMoneySummary with eventId (a bare value) and the wallet set in walletFilters. It returns one row per open market. Add a marketId filter to get a single market.
   • smartPct is smart money's Yes share of its money on the market; compare it with marketPct, the market's own probability. The gap between them is the story.
   • yesValue/noValue are dollars still held on each side, yesWallets/noWallets how many wallets, and yesPrice/noPrice smart money's average entry, not the live price.
2. Holders — unifiedSmartMoneyHolders with eventId and positionId (one side of one market), ordered by atRisk descending. Query the sibling positionId for the other side. The token lookup (query_lookup on positionId) gives each position's name and whether it is the Yes token.
3. Insider-like bets — signalScore filtered to the eventId and the wallet set, ordered by score.
4. Then ask whether the user wants to compare against another wallet set, and offer concrete options: a few relevant existing cohorts from list_cohorts, and sets that fit the event, such as the overall top wallets through userProfileLookup, a shorter duration like 1M, or a minimum roi or realizedPnl. Run the comparison with the same queries and a different filter.

On a neg-risk event (several mutually exclusive outcomes, like "Who will win?"), the No side of an outcome is smart money's bets on every other outcome, so read each market's row as "this outcome vs. the field".

Which markets smart money favors

globalSmartMoney ranks open markets by how far smart money's implied price is from the market price. Pass the wallet set in walletFilters, and to restrict it to categories, tagIdFilters with tag ids. Order by score (a 0-100 blend of edge, wallets and dollars at risk) for the strongest overall signal, or by yesSlippage for the biggest mispricing.

• side is the outcome smart money favors, smartMoneyPrice its implied price for that side and marketPrice the live one; the edge is their difference.
• Resolve marketId to the question with query_lookup on marketId, filtering id with in.
• Treat a large edge backed by only one or two wallets or little atRisk as weak.
• A negative daysLeft means the end date has passed and the market is waiting to resolve; say so rather than presenting it as a trade.

Insider signals

signalScore scores currently held positions in active markets from 0 to 100 for insider-like conviction. Useful fields: relSize (the bet against the wallet's usual size; above 1 is bigger than usual), tradeSize (peak dollars in), avgEntryPrice against pNow, and daysToResolution. Filter it by eventId, marketId, tagIds or the wallet set. Describe these as signals worth a look, never as proof of insider trading.

Presenting results

• Lead with the answer (for example "Smart money is 78% Yes vs. market 62%, $1.4M across 41 wallets"), then the table.
• Show prices in cents and money rounded ($1.2M, $45K).
• This is analytics, not financial advice; don't tell the user to place a bet.

Showing wallets

When your output is rendered as Markdown (Claude on the web and desktop, Claude Code, Codex CLI, Gemini CLI), show every wallet you mention as a Markdown link to its DataDash page: [label](https://datadash.xyz/wallet/<address>), where <address> is the wallet's address (the user id) and the label is the user's display name if known, otherwise the shortened address. For example, the wallet 0x56687bf447db6ffa42ffe2204a05edaa20f55839 is shown as Theo4 when its display name is known, and as 0x5668…5839 otherwise. Holder and signal rows return only the address; resolve display names in one query_lookup on the userId field, filtering userId with in on all the addresses you will show.