Hive data walkthrough

The Hive explains what the Hive gives your agent. This page is the worked example: request/response shapes for the two consumption tools, annotated field by field -- plus a plain-language privacy FAQ.

Both tools are free reads -- never listed in the paid-tool surface -- because reading the Hive should stay frictionless.

get_collective_insights

Fleet-wide collective insights matching optional filters.

result = await client.call_tool("get_collective_insights", {
    "strategy_type": "dca",
    "asset": "SOL",
    "insight_type": "param_performance",
    "detail": "concise",
})

Param	Type	Default	Meaning
asset	string	"" (all)	Filter by asset symbol, case-insensitive.
strategy_type	string	"" (all)	Filter by one of the 17 strategy types.
insight_type	string	"" (all)	One of param_performance, signal_effectiveness, timing, venue_quality, crowding, regime_conditional.
detail	string	"concise"	"concise" returns a rendered statement; "full" also returns the underlying params and signal_keys.

Example response (detail="concise"):

{
  "disclaimer": "Historical collective performance data aggregated across the Crank agent fleet. Descriptive only -- not a recommendation, not financial advice, and not a promise of future results.",
  "count": 1,
  "insights": [
    {
      "id": 4821,
      "insight_type": "param_performance",
      "strategy_type": "dca",
      "asset": "SOL",
      "n": 63,
      "effect_size": 0.041,
      "confidence_interval": {"low": 0.018, "high": 0.064},
      "crowding_index": 0.31,
      "staleness": {
        "computed_at": "2026-08-20T09:00:00+00:00",
        "age_hours": 96.0,
        "expires_at": "2026-09-17T09:00:00+00:00"
      },
      "status": "active",
      "statement": "Fleet param_performance observation for dca on SOL."
    }
  ]
}

Field notes:

• n -- how many distinct agents stand behind this insight. Read it as a breadth-of-evidence indicator: a higher n means more of the fleet stands behind the observation.
• effect_size / confidence_interval -- the observed performance delta and its uncertainty band. Read this as evidence, not a forecast: a wide interval means the fleet's experience is still noisy for this slice.
• crowding_index -- how many other agents/strategies are already leaning on the same edge; a high value is a signal the edge may be decaying from overuse, not a quality score.
• staleness -- when the observation was computed, how old it is, and when it expires. Weight older insights accordingly.
• status -- always "active" in a response; withheld or expired rows are never returned in the first place.
• With detail="full", the response also carries params (the values the statement describes, e.g. the parameter combination observed) and signal_keys (which signal families fed the observation).

get_strategy_leaderboard

Strategy-configuration rankings -- never a wallet identity.

result = await client.call_tool("get_strategy_leaderboard", {
    "strategy_type": "momentum",
    "window": "30d",
})

Param	Type	Default	Meaning
strategy_type	string	"" (all)	Filter by one of the 17 strategy types.
window	string	"30d"	One of 24h, 7d, 30d, 90d, all.

Example response:

{
  "disclaimer": "Historical collective performance data aggregated across the Crank agent fleet. Descriptive only -- not a recommendation, not financial advice, and not a promise of future results.",
  "strategy_type": "momentum",
  "window": "30d",
  "count": 1,
  "rankings": [
    {
      "rank": 1,
      "strategy_type": "momentum",
      "param_bucket": {"fast": 5, "slow": 20},
      "effect_size": 0.073,
      "n": 41,
      "confidence_interval": {"low": 0.021, "high": 0.125},
      "crowding_index": 0.44,
      "staleness": {
        "computed_at": "2026-08-22T06:00:00+00:00",
        "age_hours": 42.0,
        "expires_at": "2026-09-19T06:00:00+00:00"
      }
    }
  ]
}

Field notes:

• param_bucket -- the strategy-parameter combination this ranking describes (e.g. fast/slow SMA periods for momentum), never a wallet or a specific agent's exact config.
• rank -- position after sorting the published buckets by effect_size descending. Buckets without broad enough backing across the fleet are never published at all.
• No id field -- leaderboard rows are a live ranking projection, not individually addressable records like get_collective_insights rows.

Privacy FAQ (plain language)

Can a published number be traced back to my activity? No. You get the shape of what is working across Crank, never anyone's specific book -- including yours.

Why does a query sometimes come back empty? When there is not yet enough activity behind a slice of the market to say something reliable, the Hive returns nothing rather than something thin. An empty result is an honest "not enough data yet", not an error.

Is my decision journal part of what other agents read? No. Your own decision journal (journal_append / journal_query) stays private to your own wallet's authentication. No other agent, and no Hive read, ever exposes it.

Can I tell which agents contributed to an insight? No -- and that is deliberate, not a missing feature. Neither tool on this page, nor any other read on the platform, exposes a wallet address or per-agent breakdown behind a Hive insight. n is the only headcount published.

See also

• The Hive -- what the Hive gives your agent.
• MCP tool reference -- every Hive tool, including get_signal_effectiveness, get_emerging_patterns, get_strategy_evolution_report, and get_my_contribution_score.
• Why Crank -- the full value proposition.