Use CasesCrypto Hypothesis Tester — Automatically validate your market theories over time
Web 3 / CryptoResearchAutomation

Crypto Hypothesis Tester — Automatically validate your market theories over time

Recipe by @Evgeny

How it works

You define a market hypothesis once — for example "when ETH pumps 3%+, NEAR follows within 48 hours". The agent monitors prices every 6 hours, logs trigger events and their outcomes, and tracks confirmation rate over time.

Every Sunday it sends a Telegram report with how many times the hypothesis held vs failed, the current hit rate, and a verdict. After 30 days you have statistically grounded evidence whether your theory works or not — fully automated, no manual tracking needed.

Example Prompt

You are my crypto hypothesis tester. I want to test this hypothesis: "When ETH price rises more than 3% in 24h, NEAR price rises within the next 48 hours" Do the following: 1. Read memory at hypotheses/eth-near-correlation.md using memory_read. If file does not exist, create it with: # Hypothesis: ETH pump → NEAR follows within 48h - Hypothesis: When ETH rises >3% in 24h, NEAR rises within 48h - Status: TESTING - Started: [today's date] - Observations: [] - Confirmed: 0 - Refuted: 0 2. Fetch current prices from CoinGecko: https://api.coingecko.com/api/v3/simple/price?ids=ethereum,near&vs_currencies=usd&include_24hr_change=true 3. Apply hypothesis logic: - If ETH 24h change > +3%: log as TRIGGER EVENT with today's date and ETH price - If a TRIGGER EVENT was logged 48h ago: check if NEAR price is now higher than at trigger time - If YES: mark as CONFIRMED, increment confirmed counter - If NO: mark as REFUTED, increment refuted counter 4. Write updated log back to memory at hypotheses/eth-near-correlation.md 5. Every Sunday send me a Telegram summary: "🧪 Hypothesis Test Report Theory: ETH pump → NEAR follows within 48h Status: TESTING (day [X] of 30) Results so far: ✅ Confirmed: [X] times ❌ Refuted: [X] times 📊 Hit rate: [X]% Last trigger: [date] — ETH +[X]% → NEAR [outcome] Verdict: [PROMISING if >60% / INCONCLUSIVE if 40-60% / REJECTED if <40%]" Create a routine that runs every 6 hours to check prices and update the hypothesis log, and sends the weekly report every Sunday at 10:00 AM.