Crypto Reality Check

Before your agent acts on a crypto price move, a whale trade or a headline, ask one question: is it real, or noise?

Answers come from a monitor that records Binance spot and futures, Hyperliquid and 31 news sources every minute, 24/7. Each answer gives a plain verdict to act on, the numbers behind it, and a seal. The chances it states are scored in public afterwards, including when they turn out wrong.

Questions and prices

tool	Answers	Input	Price per answer
full_check	One call before acting on a token: its move, its whales and its news together, and whether they agree	token	$0.02
market_mood	What is the whole crypto market doing right now?	none	$0.002
check_move	Is this token's last-hour move its own or the market's, how unusual is it, and did moves like it keep going?	token	$0.01
whale_flow	What did the biggest traders do in this token in the last hour, by venue?	token	$0.01
check_news	Is this headline actually moving the price, or already priced in?	token and/or query	$0.01
signal_truth	Does this kind of signal work, measured on real outcomes?	optional signal, direction	$0.005

You pay only for answers. A question it cannot answer is never charged (see below).

Example

Input:

{ "tool": "check_move", "token": "BMT" }

Output (a real answer, shortened):

{
  "tool": "check_move",
  "as_of": "2026-10-23T15:34:00+05:30",
  "data_age_s": 88,
  "token": "BMT",
  "move_1h_pct": -10.03,
  "own_part_pct": -8.55,
  "market_part_pct": -1.47,
  "normal_swing_1h_pct": 1.62,
  "vs_normal_swing": 6.2,
  "verdict": { "driver": "own", "size": "extreme", "follow_through": "no_edge" },
  "history": {
    "case": "BMT 1-hour fall of 5%+",
    "episodes": 53,
    "next_hour_continued": 0.434,
    "range_99_5": [0.264, 0.621],
    "any_hour_same_direction": 0.473,
    "follow_through": "no_edge"
  },
  "chance_next_hour_continues": 0.434,
  "plain": "Mostly BMT's own move (-10.0% in 1 hour, -8.6 points its own): 6.2x its normal hourly swing, extreme. After 53 drops of 5%+ in its history, the next hour went down 43% of the time, against 47% for any hour: no edge in following it.",
  "snapshot_seal": "1f2fddf4d21c267d57813b19a8576ddffe1d2003d5d66c952b4739f0abd3ff35",
  "price_usd": 0.01,
  "charged": true
}

Every answer has verdict (short codes for logic) and plain (one or two sentences for people).

full_check returns the three answers under move, whales and news, plus verdict.overall (something_unusual or nothing_unusual) and verdict.agreement. The agreement is aligned_up, aligned_down, conflicting or no_clear_side. A price move within its normal swing takes no side.

Use it from any AI agent (MCP)

Agents on Apify's MCP server can find this Actor with search-actors and run it with call-actor, with no setup. To give your agent these tools directly, add this server to Claude, ChatGPT, Cursor or any MCP client:

{
  "mcpServers": {
    "crypto-reality-check": {
      "url": "https://mcp.apify.com?tools=zaffy_93/crypto-reality-check"
    }
  }
}

On first use the client signs in to Apify, or you can pass an Authorization: Bearer <APIFY_TOKEN> header. From code, call the Actor's API: POST https://api.apify.com/v2/acts/zaffy_93~crypto-reality-check/run-sync-get-dataset-items with the input JSON as the body. The answer comes back in the same response.

Inputs

• tool (required): one of the six above.
• token: a ticker such as BTC, ETH or SOL. BTCUSDT, $BTC and btc work too. check_move covers the top 50 Binance tokens by traded value. whale_flow also covers the coins watched for whale orders on Hyperliquid and Binance futures.
• query: for check_news, words from a headline, such as ETF approval.
• signal: for signal_truth, one of move_5m, move_1h, btc_move, volume_spike, whale_spot, whale_futures, whale_hyperliquid, liquidation, market_liquidation, funding_extreme, open_interest or news. Leave it empty for the whole table.
• direction: for signal_truth, up or down. Leave it empty for both.

When you are not charged

The run still succeeds and returns one item with "charged": false and an error saying why, when:

• the tool, token or signal is unknown;
• the market data is more than 10 minutes old (the service refuses rather than sell stale data);
• the data cannot be loaded;
• your run's spending limit is below the answer's price.

How to check it is honest

• Seals. Every answer names snapshot_seal. Each minute's market snapshot is sealed as sha256(previous seal + sha256 of the snapshot), so a past snapshot cannot be changed without breaking every seal after it. The answers are computed from that snapshot alone.
• Public record. At the top of every hour, the chance check_move states is recorded for every tracked token, whether anyone bought that answer or not, and scored an hour later. Calibration and the day-by-day seals are here: https://api.apify.com/v2/key-value-stores/Kfaru3CN4RdvD9Veg/records/scoreboard?signature=e9OKp2wi1liyUTe5e1Z1
• Failing signals are labelled failing. signal_truth reports PROVEN, PROVISIONAL, UNPROVEN or FAILED, using 99.5% bounds over independent episodes. At launch, most signals are coin flips and one reliably reverses. The service says so.

Need the raw numbers?

Crypto Market Data for AI Agents gives plain live data at $0.001 a call: prices, funding and open interest, every Binance futures liquidation, whale orders, news, Fear & Greed and order books, refreshed every minute.

Limits

• Chances are historical rates from this market's own past, not predictions or advice.
• With 50 tokens and several move sizes, a few "continues" or "reverses" labels will be luck even at 99.5%. The public record shows which ones hold up.
• Data updates once a minute. data_age_s in every answer says how old it is.
• Rates come from hourly candles. Tokens newly added to the top 50 say "no price history yet" until it is loaded.
• Public market data only: no accounts, keys or private data are involved.