/signal-to-trade-demo

name: signal-to-trade-demo
description: Prepare or run a polished BlockRun trading demo that discovers a current Polymarket market, combines live price, probability history, smart-money, and liquidity evidence into a balanced signal, produces a real order dry-run, and verifies orders or positions. Use for live demos, signal-to-trade workflows, current crypto prediction markets, or when an agent must decide whether a candidate is safe and presentable before trading.

Signal-to-Trade Demo

Run one reproducible chain: discover → verify → analyze → preview → inspect. Treat signals as evidence, never as a promise of profit. Never expose a wallet, credential, order ID, or transaction hash in presentation output.

Read [references/demo-cases.md](references/demo-cases.md) when selecting a case or preparing a fallback.

Hard safety contract

1. In a presentation dry-run, do not call wallet, setup, positions, orders, or resources. Those responses can contain wallet-derived identifiers before the final answer is redacted. The presenter performs account readiness privately before screen sharing. 2. If the current egress is blocked, never call a funds-affecting action with `confirm:true`. Continue with live data and a dry-run order preview only. A Stanford/US presentation is always dry-run mode. 3. Always preview through `blockrun_polymarket_read` `action:"preview"`. It has no confirmation input and cannot sign or submit an order. A real order requires the user's explicit approval of the exact market, outcome, amount, price/type, and current region eligibility. 4. Choose the smallest whole-dollar preview from $1–$5 that satisfies the live `min_order_size` and book depth. Never present a smaller, non-executable preview as valid. Do not split orders to bypass caps. 5. Paid market-data calls may run in parallel on Base. On Solana keep `@blockrun/llm` >= 3.8.4, which is what makes concurrent payments distinct.

1. Private operator preflight

• Confirm the Trading profile exposes eight tools and no image/video/media tool:

wallet, price, dex, markets, defi, rpc, polymarket_read, polymarket. (It was nine until 2026-09-06, when the gateway retired Surf and `blockrun_surf` went with it — a preflight that still counts nine fails.)

• Before screen sharing, the human operator may check `blockrun_wallet`, run

setup, and inspect positions/orders. Never include those raw calls in the presentation conversation.

• For the live dry-run conversation, begin directly with public market

discovery. No account state is required to preview a CLOB order.

2. Discover a current market

Use a dynamic search rather than a hard-coded condition or token ID:

blockrun_markets {
  path: "markets/search",
  params: { q: "Bitcoin", status: "open", venue: "polymarket", limit: "20" }
}

`markets/search` is the discovery path for a demo — it ranks across venues in one call. Do not automatically select the first `polymarket/crypto-updown` result because that feed can contain future placeholders with no liquidity. Rank candidates by:

• open status and a future close time;
• an unambiguous resolution source and threshold;
• non-zero 24-hour volume and trade count;
• a valid condition ID plus outcome token IDs;
• an outcome price away from 0 and 1;
• a live order preview that finds a usable book.

Resolve the selected market using `polymarket/markets/keyset` with `condition_id`, `status:"open"`, and a small `limit`. Do not invent Gamma-only parameters such as `active`, `closed`, `order`, or `ascending`; the MCP rejects those before payment. Predexon's own `search`, `sort`, `end_after`, and `end_before` filters are supported on that endpoint.

3. Collect evidence sequentially

Use four independent lenses where the market supports them:

1. **Underlying:** get current BTC/USD with `blockrun_price`, then compute the exact percentage move required to reach the market threshold before expiry. 2. **Probability trend:** query the selected Yes token:

   blockrun_markets {
     path: "polymarket/candlesticks/token/<TOKEN_ID>",
     params: { interval: "1440", start_time: "<UNIX_SECONDS>", end_time: "<UNIX_SECONDS>" }
   }

`interval` is integer minutes (`1440`, not `1h`) and is optional. `60` was observed returning a paid 400 where `1440` worked; `start_time` and `end_time` are Unix seconds. 3. **Smart money:** use a meaningful cohort:

   blockrun_markets {
     path: "polymarket/market/<CONDITION_ID>/smart-money",
     params: { window: "30d", min_trades: "100" }
   }

Report wallet count, net-buyer share, volume, and aggregate PnL. A high buyer share with negative PnL is mixed evidence, not automatically bullish. 4. **Liquidity/history:** query historical orderbooks with `token_id`, `start_time`, and `end_time` in Unix milliseconds. The order dry-run is the authoritative live fillability check.

Record the timestamp and data source for every observation. If a source fails, label it unavailable and continue; never manufacture a value.

4. Build the signal

Present an evidence table with these columns:

| Source | Observation | Supports | Reliability | |---|---|---|---| | Spot vs threshold | Exact distance and time remaining | Yes/No/Mixed | High | | Market trend | Probability change over a fixed window | Yes/No/Mixed | Medium | | Smart-money cohort | Buyer share, volume, PnL | Yes/No/Mixed | Medium | | Book/liquidity | Spread, available size, 24h activity | Executable/Thin | High |

Then state:

• one-sentence thesis;
• strongest counterevidence;
• confidence (`low`, `medium`, or `high`) with a reason;
• proposed side and the smallest executable whole-dollar preview from $1–$5,

or `NO TRADE` when gates fail.

Do not describe the result as financial advice or a guaranteed “good signal.”

5. Preview and verify

Preview through the dedicated non-destructive action:

blockrun_polymarket_read {
  ac