Synthetic data & provenance

David is a synthetic data API. Understanding exactly what is generated versus what is real is essential to using it correctly.

​

What is synthetic

Effectively everything that carries financial meaning is generated:

• Prices and OHLCV bars
• Income statements, balance sheets, cash-flow statements, and metrics
• Earnings results, surprises, guidance, and analyst estimates
• News articles, analyst notes, and SEC-style filings
• Insider trades, institutional holdings, and corporate actions
• The macro tape and central-bank policy path
• Events and their effects

None of these are real issuer histories. They are deterministic functions of a scenario’s seed and configuration.

​

What is real

Two things are grounded in real-world reference data:

1. Ticker symbols. David uses real exchange-style symbol aliases (e.g. AAPL, NVDA, SPY) so your tooling and ticker lists work unchanged. The bundled universe contains 16,143 unique public US symbols from SEC/Nasdaq snapshots.
2. Company reference identity. Each company’s display metadata (name, sector, industry, business description, market-cap tier, instrument type) reflects the real-world company, so AAPL reads as Apple, Consumer Electronics and SPY reads as an S&P 500 ETF.

​

Determinism

David’s defining property is determinism. A scenario is built from a seed plus configuration, and the same inputs always produce the same world, down to individual price bars and headline text. Generation also follows a fixed dependency order, so the world is internally consistent rather than a pile of independent random series: David first generates the hidden truth of the world, then derives fundamentals, expectations, events, prices, and finally the public artifacts from that truth. The hidden state is never exposed through the API; only its downstream, point-in-time observable effects are.

​

Internal consistency

Because the world is derived from a single hidden truth, the data ties together:

• Accounting identities hold (balance sheet balances, cash reconciles, EPS ties out).
• OHLC and volume invariants hold (low ≤ open, close ≤ high; event days spike volume).
• Prices react to earnings: significant beats/misses move the stock in the right direction.
• News and filings repeat the numbers in the structured records instead of drifting from them.
• Filings carry the metadata (accession, CIK, item codes) and link back to the events that triggered them via source_event_id.

Every scenario ships a validation report that checks these properties.

​

Calibration

David’s generator is tuned against stylized facts of real markets (return distributions, volatility clustering, drawdown frequency, volume/reaction coupling, tail behavior, and cross-sector correlation) measured from market panels. The /metadata/empirical-calibration endpoint reports the current calibration manifest and gate margins.

​

Next steps