Post-mortem · Case study

Why we stopped trading bots

For 10 months we ran automated Kraken bots on real money. They won 73% of individual trades — and still lost 63 percentage points against passive BTC buy-and-hold. Here is the full data set, what we learned, and why Swiss Whale Intelligence is now an information service, not a trading product.

Period: July 2025 — April 2026  ·  Trades: 682  ·  Reproducible: SHA256-hashed CSV

📊 The headline number

Compound equity of the Kraken bots vs. passively holding BTC over the same period, computed directly from the published CSV. Switch between all-time and the last 90 days.

View:
SWI compound P&L
BTC buy-and-hold same period
Δ active management cost
win rate

🧠 Why this happened (spoiler: fees + time off the market)

The bots bought and sold spot BTC. That means each individual trade tracked the BTC price perfectly — per-trade alpha vs. BTC in the same hold window was effectively zero. So the losses are not "bad picks". They are structural:

🎯 The trust-score paradox

We also stress-tested one of the core assumptions — that "high trust" wallets (exchanges, institutions, labelled whales) are better signals. In our data they are the worst signals.

Trust bin N 6h return Win rate
New (<0.1)342+0.438%56.1%
Low (0.1–0.3)43+0.340%60.5%
Mid (0.3–0.8)952+0.136%56.4%
High (0.8+)383−0.089%46.5%

Interpretation: trust correlates with being known, not with being profitable. Labelled institutional flow is already priced in by the time it hits the public mempool. The signal — if there is any — comes from new wallets that nobody has labelled yet.

🧬 What we changed as a result

The 2025 experiment killed the "trading bot" hypothesis for us. We kept the infrastructure we had built (local Bitcoin full-node, mempool watcher, 835K watchlist addresses, 711 hand-curated entity labels, 7.68M indexed whale transactions) and repurposed it as an information product:

📁 The raw data

Every number above is derived from these two CSV files. Schema is stable; new rows append, existing rows never change.

📥 latest_all.csv (full period) 🔐 SHA256
📥 latest_90d.csv (rolling) 🔐 SHA256
📁 Browse weekly archive 📄 metadata.json

🔒 Verify a download

Pick the CSV you just downloaded — your browser computes SHA256 locally (no upload) and compares to the published hash.

Command-line equivalent:

curl -s https://btcwhalealerts.com/ledger/latest_all.csv | openssl dgst -sha256
curl -s https://btcwhalealerts.com/ledger/latest_all.csv.sha256

⚠️ Caveats

📨 Questions?

Academic, journalistic or research use is welcome. The data is openly licensable for citation. See our imprint for contact details.