Whales splitting orders to hide their hand.
IT'S ALPHA
Pure research & development. This page exists so we can test whether the on-chain data patterns are coherent enough to derive insight at the boundaries of what's observable. Methodology is disclosed, error bounds are surfaced where possible, but accuracy is not guaranteed and we may rewrite, replace, or remove anything you see here at any time. Do not trade or make decisions on this output without independent verification.
What is alpha: read the full alpha protocol →
Live detections heuristic
Same sender · ≥5 privacy txs · within 90s · clustered cipher size
Scanning privacy buffer…
What we're testing
Some whales avoid market impact by splitting one large order (say $5M notional) into N smaller privacy-mode children fired in a tight time window. Private Mirror aggregates privacy traffic across the chain. Iceberg Hunter focuses on the intra-wallet pattern: same sender + N privacy orders within W seconds + size-class clustering = iceberg signal.
We never decrypt. The detection signal lives entirely in metadata we can already see (sender, timestamp, action type, position-in-block). Calibration vs a labelled subset gives a per-wallet false-positive rate.
Methodology (sketch)
- For each (sender, window=W) slice, count privacy-mode PlaceOrder txs.
- Iceberg score increases with: (a) count above threshold, (b) inter-arrival time variance below threshold, (c) sender historical privacy-share above 50%.
- Confidence band = bootstrap on the labelled training set. Reported as 95% CI half-width, never as a point estimate alone.
- Output: live feed (last 6h) + per-wallet score + evidence trail (the contributing tx hashes, even though their payloads are encrypted).
Labelled subset · 1 positive
Ground truthHand-confirmed iceberg-pattern wallets with full evidence. The calibration pipeline bootstraps from this set: any candidate scored by the live detector is benchmarked against these positives to compute the per-wallet false-positive rate.
| Address | Tx/s | Window |
|---|---|---|
| 0x6DCe…c6f4 LAB-USDT MM bot | 5.08 | 59s |
Coming in v0.2
- False-positive rate calibration vs the labelled subset above
- Inter-arrival variance signal — tighter clustering raises confidence
- Cancel-ratio anti-pattern — MMs cancel a lot, real icebergs don't
- Push to Telegram alerts when a high-confidence iceberg fires
Want to suggest an iceberg-pattern wallet? DM @aster_scan.