Methods and limitations

All figures in reports are calculated by code from primary sources, stored in a local SQLite database, and can be audited. The language model only writes narrative from those figures.

Data sources

Prices and market cap	Public CoinGecko API, top 60 assets, snapshot every 15 minutes. 400-day daily history from Binance klines for assets listed on Binance.
Derivatives	Binance USD-M Futures: 8-hour funding rate, open interest, account long/short ratio, taker buy/sell ratio. Only three assets (BTC, ETH, SOL) to preserve quota.
Macro	S&P 500 and VIX from CBOE (daily CSV). US Treasury yield curve (2Y, 10Y, 30Y). EUR/USD, USD/JPY, USD/CNY, USD/GBP rates from ECB. Gold uses PAXG from CoinGecko.
Liquidity	Aggregate stablecoin supply and DeFi TVL from DefiLlama. US spot ETF flows from Farside Investors (read daily). Bitcoin network statistics from Blockchair.
Sentiment	Fear & Greed Index from Alternative.me, 7- and 30-day averages, plus one-year percentile.
News	RSS from CoinDesk, Cointelegraph, The Block, Blockworks, Decrypt, Federal Reserve, ECB, BIS, and Google News queries for macro, geopolitics, regulation, and ETFs.

Metric definitions

Regime	Composite score from VIX, S&P 500 position relative to MA50, 20-day change in UST 10Y, BTC position relative to MA200, and 7-day breadth. Score -3 or lower = risk-off.
Funding annualized	8-hour funding multiplied by 3 and by 365. Positive means longs pay shorts, a sign that long positioning is crowded.
1-year percentile	Current value's position within the last 365-day range. 90 means the current value is higher than 90% of days in the year.
Breadth	Number of assets with positive 7-day change, and number of assets above MA50 from the sample with sufficient history.
Drawdown	Distance of current price from the 400-day peak, in negative percent.
Confidence	Editorial assessment of evidence quality: high when there are many independent sources and long data history, low when it relies on a single series or short sample.

Limitations to be aware of

ETF data is available only for trading days and has been collected since October 2026, so long-term flow analysis is not yet possible.
Derivatives are only from Binance, not an aggregate across all exchanges. Positions on other exchanges are not visible.
CoinGecko rate-limits requests from VPS IPs. When the limit is hit, the collector uses the last cache and notes it in the log.
Report narratives are composed by a language model from numbers calculated by code. The model never invents numbers, but emphasis and interpretation can be wrong and should be read critically.
There is no institutional positioning data (13F), paid on-chain exchange reserve data, or options data at strike level. Positioning analysis is therefore incomplete.

Workflow

Collector runs every 15 minutes (prices, derivatives) and every 30 minutes (news). Daily data and macro are pulled once a day.
Analysis calculates metrics and signals with fixed rules in analyze.py.
The report writer composes the narrative from the fact sheet, then the numbers are rechecked against the database.
Watcher monitors significant thresholds every 20 minutes and only speaks up when something important happens.
Research hypotheses are recorded, tested, and the results are shown openly on the hypotheses page.

Updated 25 Oct 2026 06:17 UTC+7. Not investment advice.