Task Card

A task card is the structured research work order used by Factor Mining.

A task card tells the agent what market behavior to investigate, which data headers it may use, what forward horizon applies, and which research context should shape the factor.

A research task is the topic.
A task card is the instruction packet.

Public task lists normally emphasize the human-readable task name and canonical category. Internal task handles are opaque selectors: the agent should use an exact returned handle and never guess, edit, or derive one.

Why Task Cards Exist

A vague request such as find a profitable trading idea is too broad to evaluate responsibly. A task card narrows the job to a market mechanism, supported data, and a declared test horizon.

For example:

Investigate whether deteriorating liquidity precedes market fragility.
Use only the task's allowed headers.
Generate contract-compliant plugin.py and a readable formula.
Validate the complete source before submission.

Task Information You May See

Field	Meaning
name	Human-readable task name
category	Canonical family such as Microstructure, Technical, or Volatility
description	The market behavior to investigate
allowed data	Exact headers available to the factor contract
forward period	Evaluation horizon in bars; current public tasks use seven daily bars
status	Whether the task is currently open
core question	The precise research question
primary alpha source	The proposed source of predictive information
economic principle	Why the mechanism could exist
microstructure or crypto mechanism	Market-specific explanation when applicable
research directions and feature hints	Candidate approaches, not required formulas
regime considerations and risk sources	Conditions that may weaken or invalidate the idea
target behavior	What successful factor output should capture

The exact returned object can contain additional structured context. Use that object and the scoped factor contract as the authority instead of copying a static example from this page.

Simplified Example

{
  "name": "Market Microstructure And Liquidity Fragility",
  "category": "Microstructure",
  "description": "Test whether liquidity deterioration predicts market fragility.",
  "core_question": [
    "Does weakening liquidity precede unstable price behavior?"
  ],
  "allowed_data": [
    "open",
    "high",
    "low",
    "close",
    "volume",
    "quote_volume",
    "taker_buy_volume",
    "taker_sell_volume",
    "open_interest_close",
    "funding_rate_close",
    "liquidation_long_usd",
    "liquidation_short_usd",
    "binance_premium_index_close"
  ],
  "fwd_period": 7,
  "status": "open"
}

This example is explanatory. The task and contract returned in the current session decide the exact fields, data headers, and horizon.

How The Agent Uses A Task Card

Good agent behavior:

• select one exact returned task;
• restate its objective and mechanism;
• read the scoped construction contract;
• use only exact allowed headers;
• check for materially similar existing work;
• generate and fully validate plugin.py;
• explain assumptions, applicability, and risks;
• ask before submitting a mutation.

The task card is a constraint, not a suggestion.