danjsiegel/whaledoxer-alpha

🔱 WhaleDoxer Alpha: Insider Signals & Audit Trail

WhaleDoxer is a high-precision surveillance engine for prediction markets (Polymarket & Kalshi). This dataset contains an audit trail of Anomalous Trades that were flagged by the engine's Z-Score sensors.

This dataset is flattened to include both Resolved trades (where the final outcome is known) and Open positions (live anomalies still pending market settlement). You can filter these using the status column.

🔬 Dataset Purpose

This data is intended for researchers and data engineers building Information Asymmetry Models. By analyzing the "footprints" of these anomalous trades against their actual outcomes (Brier Scores) and returns (ROI), you can identify patterns associated with "Informed Flow" (Insiders) versus "Speculative Noise" (Gamblers).

🛡️ Anonymization & Privacy

To protect market participants while preserving behavioral relationships, this dataset has been sanitized:

• Hashed Entities: All 0x wallet addresses have been SHA-256 Hashed with a private salt (sha256(lowercase(wallet_address) + salt)). Researchers can track the same entity across multiple trades without unmasking them.
• Unresolved / Anonymous Trades: If a trade was executed by an anonymous user (e.g., Kalshi volume) or an unresolved Polygon proxy, the wallet_hash is left as null to prevent false entity clustering.
• PII Scrubbing: Plain-text usernames detected in the scorer_breakdown reasoning logs have been automatically scrubbed and replaced with "REDACTED" via regex.

📊 Schema Description

Column	Type	Description
wallet_hash	string	SHA-256 anonymized entity identifier.
trace_id	string	Unique UUID for the paper trade evaluation.
market_id	string	Unique ticker (Kalshi) or ConditionID (Polymarket).
market_title	string	Human-readable title from the market catalog.
source	string	Exchange source (kalshi or polymarket).
direction	string	Position direction (e.g., BUY, SELL).
trade_timestamp	timestamp	When the triggering anomalous trade hit the exchange ledger.
z_score	float	Statistical deviation of the trigger trade from the rolling volume baseline.
trigger_price	float	Execution price of the anomaly that tripped the wire.
trigger_amount	float	Contract volume of the anomaly that tripped the wire.
size_usd	float	Notional value in USD of the paper trade position.
entry_price	float	Actual price WhaleDoxer logged the paper trade at.
exit_price	float	Final settlement payout price (1.0 for WIN, 0.0 for LOSS) (Null if open).
p_insider	float	Composite suspicion score calculated by WhaleDoxer.
brier_score	float	Accuracy metric: (p_insider - outcome)^2. Lower is better (Null if open). I did f up the actions (which I did for a while), and did the inverse 🤦
outcome	float	Final settlement result (1.0 for YES, 0.0 for NO) (Null if open).
roi	float	Return on investment for the paper position (Null if open).
is_real	boolean	Flag indicating if this was executed on real capital vs paper.
trigger_latency_ms	float	Total glass-to-glass latency of the forensics pipeline.
market_close_time	timestamp	Expected resolution time for the market.
created_at	timestamp	When WhaleDoxer logged the paper trade.
resolved_at	timestamp	When the market was officially settled (Null if open).
status	string	Either resolved or open.
scorer_breakdown	variant (JSON)	Nested JSON containing plugin sub-scores and redacted reasoning strings.

🚀 Usage

This dataset is exported weekly via the WhaleDoxer Airflow Pipeline. It is stored in ZSTD-compressed Parquet format for maximum query performance.

from datasets import load_dataset

dataset = load_dataset("dansiegel/whaledoxer-alpha")
df = dataset['train'].to_pandas()

# Filter for resolved trades only
resolved_df = df[df['status'] == 'resolved']