danjsiegel/whaledoxer-alpha

WhaleDoxer Alpha: Insider Signals and Audit Trail

This dataset contains the WhaleDoxer paper-trade audit trail exported from PostgreSQL. For more information what this dataset is and why check out the substack article I wrote: https://danjsiegel.substack.com/p/building-a-sub-second-middle-finger

Each row is a paper trade written by the downstream forensics pipeline after a tripwire was evaluated and the write gate was passed. The export includes both:

• resolved trades, where outcome and evaluation fields are known
• open trades, where the market has not yet settled

Use the status column to distinguish them.

What This Dataset Contains

This is not a dump of all raw trades or all tripwires.

It is a weekly export of rows from paper_trades, joined back to the latest matching tripwire row and market catalog metadata. In practice, that means this dataset contains the subset of detected anomalies that were strong enough to become paper trades.

Export Details

• Export cadence: weekly
• Storage format: Parquet with Zstandard compression
• File pattern:
  • data/paper_trades_YYYY-MM-DD.parquet

Privacy and Anonymization

Wallet hashing is performed in Python before export:

• wallet_hash = sha256(lower(wallet) + salt)

The following values are emitted as null instead of being hashed:

• empty strings
• anonymous_lookup_required
• AMM_Liquidity_Pool
• values starting with kalshi::
• any non-0x identifier

In addition, the export applies a narrow regex redaction pass to scorer_breakdown_json, replacing text that matches:

• Wallet .*? entered

with:

• REDACTED

That redaction is limited to the exported scorer-breakdown text field. It is not a general-purpose scrubber for arbitrary prose.

Schema

Column	Type	Description
wallet_hash	string | null	SHA-256 hash of a lowercased on-chain wallet address. Null for sentinels and non-0x identities.
trace_id	string | null	Trace identifier for the paper-trade evaluation path.
market_id	string	Exchange market identifier: Kalshi ticker or Polymarket condition ID.
market_title	string | null	Human-readable title from the market catalog when available.
source	string	Exchange source, taken from the joined tripwire when present and otherwise inferred from market_id.
direction	string	Paper position side: yes or no.
trade_timestamp	timestamp | null	Timestamp of the triggering anomalous trade from the joined tripwire row.
z_score	double | null	Z-score of the triggering trade from the joined tripwire row.
trigger_price	double | null	Price of the triggering anomaly from the joined tripwire row.
trigger_amount	double | null	Size of the triggering anomaly from the joined tripwire row.
size_usd	double | null	Notional USD size recorded for the paper trade.
entry_price	double	Entry price recorded when the paper trade was written.
exit_price	double | null	Settlement price written after resolution. For binary markets this mirrors the resolved outcome price. Null if open.
p_insider	double	Composite suspicion score written by the forensics pipeline.
brier_score	double | null	Direction-aware Brier score written after resolution. Null if open.
outcome	double | null	Resolved market outcome encoded as 1.0 for YES and 0.0 for NO. Null if open.
roi	double | null	Direction-aware binary-market return written after resolution. Null if open.
is_real	boolean | null	Whether the recorded trade was marked as real-capital execution versus paper/simulated.
trigger_latency_ms	double | null	End-to-end latency from triggering trade timestamp to paper-trade write time.
market_close_time	timestamp | null	Expected market close or resolution time recorded at paper-trade creation.
created_at	timestamp	Timestamp when WhaleDoxer wrote the paper trade row.
resolved_at	timestamp | null	Timestamp when the market was settled and the row was marked resolved.
status	string	resolved if resolved_at is present, otherwise open.
scorer_breakdown_json	string | null	JSON text representation of the scorer breakdown exported from PostgreSQL JSONB.

Evaluation Semantics

For resolved rows, the export includes evaluation fields computed downstream.

Direction semantics:

• direction = "yes" means the paper position was on YES
• direction = "no" means the paper position was on NO

Brier score semantics:

• if direction == "yes", predicted probability is entry_price
• if direction == "no", predicted probability is 1 - entry_price
• brier_score = (predicted_probability - outcome) ** 2

ROI semantics:

• if direction == "yes", roi = outcome - entry_price
• if direction == "no", roi = entry_price - outcome

Notes

• Open rows will have nulls in exit_price, brier_score, outcome, roi, and resolved_at.
• scorer_breakdown_json is exported as JSON text in Parquet, not as a nested VARIANT type.
• This dataset is best treated as an audit trail of paper-trade decisions, not as a complete sample of all anomalous exchange activity.

Loading Example

from datasets import load_dataset

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

resolved = df[df["status"] == "resolved"]
open_positions = df[df["status"] == "open"]