Skip to main content

​
Prefer url over text for structured data sources

URLs pointing to data pages (government statistics, financial reports, research databases) tend to yield more accurate and complete timeseries than pasted text. 
// Preferred for structured data
{"url": "https://data.worldbank.org/indicator/NY.GDP.MKTP.CD?locations=US"}

// Use text for unstructured content like articles or PDFs
{"text": "Revenue in 2021 was $5.2B, growing to $6.8B in 2022 and $8.1B in 2023."}

​
Include units in text input

If submitting raw text, including the unit makes extraction more accurate. 
// Good — unit is explicit
{"text": "US CPI (index, 2015=100): 2020: 258.8, 2021: 271.0, 2022: 292.7"}

// Less reliable — no unit context
{"text": "2020: 258.8, 2021: 271.0, 2022: 292.7"}

​
Validate timestamps

Timestamps may be returned as ISO 8601 dates or human-readable strings depending on the source. Normalize them before rendering. 
const normalized = dataPoints.map(p => ({
  ...p,
  timestamp: new Date(p.timestamp).toISOString(),
}));

​
Combine with Causality Graph

After extracting a timeseries (e.g., rising inflation), use Causality Graph to understand the causal drivers behind the trend. 
Timeseries (trend data) → Causality Graph (why did this happen?)