danjsiegel/whaledoxer-stream

🐋 WhaleDoxer Telemetry Stream

This dataset contains live, rolling telemetry from the WhaleDoxer Forensics Engine, an algorithmic surveillance system monitoring prediction markets (Polymarket and Kalshi) for insider trading and informed whale aggression.

Data is appended incrementally every 6 hours as Parquet files, retaining the native VARIANT JSON schema for sub-field analysis.

⚡ Performance Profile (The "True Claim")

WhaleDoxer is built on a decoupled Go/DuckDB/Local-LLM architecture, running entirely on edge hardware (Apple Silicon). Our latency profile is tiered:

• L0 Ingestion (< 10ms): WebSocket byte frame receipt and normalization.
• L1 Statistical Trigger (< 150ms): In-memory DuckDB calculates 30-day moving averages and standard deviations, firing when volume crosses a 4.0σ (Kalshi) or 3.0σ (Polymarket) threshold.
• L2 Forensic Enrichment (~1.5s): The engine resolves anonymous on-chain Polygon addresses to human identities and queries a local LLM to generate a probabilistic conviction score (p_insider).

📊 Schema

Column	Type	Description
wallet_hash	string	SHA-256 hashed wallet address (anonymized).
event_timestamp	timestamp	Exchange-reported trade time (the trigger event).
source	string	Exchange source (polymarket or kalshi).
market_id	string	The unique condition ID or ticker.
market_title	string	Human-readable market question.
z_score	double	Volume anomaly depth at the exact time of the fire.
sniper_latency_ms	double	Milliseconds from WS receipt to DB write.
p_insider	double	Composite AI conviction score (0.0 to 1.0). NULL if rejected.
trigger_latency_ms	double	Total glass-to-glass latency (whale trade to final forensics write).
scorer_breakdown	variant	Native DuckDB JSON containing per-plugin subsystem scores (topic_reasoning, keyword_st, wallet_su, etc.).