Methodology

Read the Tape reads financial news and tracks how the conversation around each asset evolves — which themes are gaining attention, whether coverage tone and underlying arguments agree, and how those narratives shift day to day. It is a lens on market discourse, not a price model.

What we cover

Twelve assets across crypto, equities, and commodities/rates, refreshed every day:

  • Bitcoin
  • Ethereum
  • Solana
  • S&P 500
  • Nasdaq 100
  • Dow Jones
  • Russell 2000
  • NVIDIA
  • Google
  • US 10Y Treasury
  • Gold
  • Silver

How it works

  1. Collection. Each day we gather roughly 250 recent articles about these assets from public news and market sources.
  2. Extraction. Large language models (Anthropic Claude) read each article to identify the themes it advances, the direction of the argument (bullish / bearish / neutral), and the tone of the language.
  3. Clustering. Semantically similar claims are grouped into canonical narratives using text embeddings, so the same thesis expressed in different words is counted as one theme and tracked over time.
  4. Scoring. We compute a small set of metrics for each asset and theme — sentiment, narrative balance, coverage tension, volume share, and momentum — and assign each theme a lifecycle stage. Every term is defined on the Definitions page.
  5. Synthesis. A daily language-model pass writes the plain-English summaries you read — what sources are saying and why a theme matters — grounded in the extracted evidence.

Cadence & dates

The pipeline runs once nightly, shortly after midnight U.S. Pacific time, and covers the prior day's coverage. Each page is stamped with the data date it reflects (as of <date>). Prices are the latest available quote from market-data providers (CoinGecko for crypto, Yahoo Finance for equities and commodities) as of that date — not live streaming quotes.

Limitations

We are candid about what this is and isn't:

  • It is AI-generated and imperfect. Automated reading can misjudge sarcasm, satire, or context, and can occasionally miscategorize a theme.
  • Coverage varies by asset. On thin news days some assets have too few articles to score; those are shown as an explicit low- or no-coverage state rather than a confident-looking number.
  • It measures the conversation, not the market. Sentiment and narrative signals describe what coverage is saying — they are not price forecasts, and the conversation can be wrong or already priced in.
  • It is not financial advice. Nothing here is a recommendation to buy, sell, or hold any asset.

Questions & corrections

Read the Tape is built by Spearman Analytics. If you spot an error or have a question about how something is computed, we want to hear about it — reach out through the Spearman Analytics site.