Methodology

Market Mountain aggregates financial news from six independent sources running in parallel: Finnhub, NewsAPI, curated RSS feeds (Bloomberg, Financial Times, CNBC, Reuters, WSJ, and 20+ others), Marketaux, NewsData, and GNews. Each source operates independently — if one fails, the others continue without interruption. Articles are filtered by age (48-hour window), relevance (217 financial keyword matches), and source quality (domain blocklist removes entertainment, tabloid, and PR wire outlets).

Raw articles are grouped by topic using keyword clustering and entity extraction. Groups with 2+ independent sources are prioritized over single-source stories. A multi-layer deduplication system prevents the same event from generating multiple articles: cross-run topic cooldowns (8-24 hours depending on category), entity-based event matching with canonical synonym normalization (e.g., "crude oil" and "WTI" map to the same entity), and headline similarity scoring with both raw and paraphrased matching.

Each qualified topic group is synthesized into a single editorial article using Claude (Anthropic). The synthesis prompt enforces a Bloomberg/Financial Times editorial tone, structured output (headline, key takeaways, why it matters, second-order implications, what to watch), and single-story focus. Market data from FRED, BLS, EIA, and FMP is provided as context so Claude can ground claims in real numbers rather than hallucinating statistics.

Every synthesized article passes through four verification layers before publication:

Layer 1 — Heuristic plausibility scoring evaluates whether claims contain specific financial data (percentages, dollar amounts, basis points) and source attribution. Unsourced assertions score below the threshold and are rejected.

Layer 2 — Data verification cross-references 14 claim patterns against live government and market data from FRED, BLS, EIA, and FMP. Verified claims include Fed funds rate, 10-year yield, CPI, unemployment, payrolls, WTI crude, S&P 500, VIX, dollar index, GDP growth, mortgage rates, and gold prices. Each has calibrated tolerance bands (e.g., CPI within 0.2 percentage points, payrolls within 15% of claimed value).

Layer 3 — Source alignment uses AI to verify that every synthesized claim traces back to at least one source article. Claims that cannot be grounded are flagged as potential hallucinations.

Layer 4 — Entity relationship verification detects fabricated relationships (e.g., false M&A claims). Any fabricated relationship caps the composite score at 30/100, triggering automatic rejection.

Articles that pass fact-checking face an 18-test editorial quality assessment scoring 0-100. Tests include story worthiness, source quality (Tier 1/Tier 2 classification), title quality, thesis clarity, section structure, key takeaways, chart data quality, originality vs. recent coverage, editorial voice (detecting generic filler phrases), and data verification scores. The production threshold is 90/100 — articles must score A-grade across nearly every dimension to publish.

Sources are classified into tiers based on editorial standards. Tier 1 includes Reuters, Bloomberg, Associated Press, CNBC, Financial Times, Wall Street Journal, New York Times, The Economist, and Barron's. Tier 2 includes Yahoo Finance, MarketWatch, Seeking Alpha, Business Insider, Forbes, and CNN Business. Untiered sources receive lower confidence scores. Tier 1 sources receive a verification badge on the site.

Long-form equity research uses multiple valuation models that must converge to support a price target. Primary methods include discounted cash flow (DCF) analysis with sensitivity tables varying WACC and long-term growth rate, EV/EBITDA multiples vs. peer group, P/E ratio analysis, and price-to-cash-flow comparisons. All articles include a DCF heatmap showing the range of fair values under different assumptions, explicit risk factors, and a clear investment thesis. Price targets and their performance are tracked on the Track Record page with live market prices.

The Daily Markets Briefing is automatically generated each trading day at 8:00 AM Eastern. It synthesizes the day's published stories into a lead story summary, top developments, key macroeconomic data (Fed funds, 10-year yield, 2s-10s spread, CPI, WTI crude, dollar index from live sources), and three forward-looking "What to Watch" items prioritized by the FMP economic calendar (FOMC decisions, CPI releases, NFP reports, GDP prints). The briefing maintains multi-day continuity — it references yesterday's watch items and generates follow-ups when events resolve.