Quantitative Signal System

The NataPulse quantitative system transforms canonical market candles into structured probabilistic evidence.

It is deliberately separated from the public event stream. Not every model output becomes a NataPulse event.

Pipeline

Market providers
  → canonical OHLCV candles
  → candle-quality snapshot
  → quality gate
  → quantitative inference
  → stored quant signal
  → selective promotion policy
  → optional market_quant event
  → cluster, report, or research evidence

Signal types

The current public model supports:

Forecast: a hedged bullish, neutral, or bearish bias with strength.
Volatility: expected volatility evidence relative to the evaluated window.
Anomaly: a score indicating unusual short-term behavior.

Signals can also expose timeframe, lookback or horizon, confidence, generation time, data quality, source-data references, and model provenance where product-safe.

Quality gate

Inference requires a usable candle window. Missing, duplicate, stale, incomplete, or invalid data can suppress generation or mark the result as degraded.

A strong-looking signal with poor input quality should not be treated as strong evidence.

Promotion policy

A signal becomes a public market_quant event only when the promotion policy finds sufficient evidence. Conceptually, promotion requires either:

at least one strong trigger; or
multiple weaker, mutually supporting triggers.

Strong triggers can include a high anomaly, a substantial volatility spike, or agreement across timeframes. Weaker triggers can include an elevated anomaly, moderate volatility change, a strong non-neutral forecast, or market-volume confirmation.

One weak directional forecast is not enough by itself.

Cluster behavior

A promoted quant event can attach to an existing active cluster for the same asset as quantitative evidence. A single quant signal does not create a persistent cluster by itself.

Reports and research

Valid quant signals can appear in reports and Deep Research. When no other source confirms them, the system marks the evidence as unconfirmed and avoids using it as the sole basis for a directional conclusion.

Quant Signals product surface

The dedicated Quant Signals terminal presents the product-safe coverage view rather than the raw model store. It can combine:

the covered portion of the supported equity universe;
forecast-bias distribution;
anomaly counts and sector context;
a Signals view for names with available model evidence;
an Index view that keeps uncovered names visible as an honest no-signal state;
an inspector that separates forecast, volatility, anomaly, confidence, and current market context.

The standalone terminal does not display a per-signal data-quality status. Quality gating happens upstream; use the generated and active or expired states for visible timing context. When a signal is promoted into an event, Event Explorer can expose its quality status when available.

Coverage is intentionally not inferred for an asset with no row. A missing signal and a neutral signal are different states.