Research note 002: the tape is dust plus whales

Date: 2026-10-10 · Status: observational, single 30-minute window. Companion to note 001 (count vs volume). Same honesty rules: no profitability claim, limits stated, method reproducible.

Question

What does the size distribution of taker prints imply for signal design? Note 001 showed count and volume disagree; this note measures how the volume itself is distributed.

Dataset

• 14,979 trade prints, Binance spot BTC/USDT, 30 minutes.
• One connection gap mid-capture (venue_close, single attempt), recorded inline; audit verdict healthy, clean duration stop. The gap is part of the evidence: the pipeline met reality once and marked it.
• Normalized to one Parquet file (14,979 exact-decimal rows).

Results

Overall: 7,911 buys (52.8%) vs 7,068 sells; 64.744 vs 62.826 BTC by volume (50.8% buys). Per-minute buy shares swing 7.7%–94.5% by count — the note-001 pattern persists over a 3x window.

Size distribution (BTC per print):

The top 1% of prints by size carries 46.4% of total volume.

Reading

Half the prints are dust: the median print is ~$8, on which a 5 bps taker fee is under half a cent — individually executable, informationally worthless. Nearly half the volume arrives in 1% of prints. Any signal that weights prints equally (counts, or unweighted averages) lets dust outvote size by orders of magnitude; any signal that could matter must weight by size or filter by it, and then faces the note-001 problem (size shows no persistent side). Between dust that says nothing and size that says nothing consistently, the per-print tape offers no tradeable signal in this window — the third consecutive "no," each for a different measured reason.

What would strengthen or overturn this: size-conditioned side persistence (do large prints lead short-horizon drift? — needs book context per print, not available in trade-only captures); multi-venue size distributions; fee-floor analysis per print-size bucket through the fill model.