Nourrétance processes continuous market feeds, isolates statistically relevant signals and returns actionable confidence scores, with a processing latency of less than 80 milliseconds.
Per-asset trust score, instantaneous network latency, aggregated risk exposure, and historical signals — updated every 400 milliseconds.
Quote flows, volumes and macroeconomic indicators are collected in real time, without delayed sampling.
The series are adjusted for seasonality and anomalies, to limit false signals from intraday volatility.
An ensemble of models evaluates multiple scenarios and produces a probability distribution, rather than a single, deterministic prediction.
The confidence score and its explanatory variables are transmitted to the dashboard or, if configured, to an automated execution module.
These values correspond to standard processing in a production environment and may vary depending on the volume of the flow analyzed.
The objective is not to predict the market with certainty, but to reduce the part of decisions taken under the effect of emotional bias and informational noise.
The risk aggregated by position and by portfolio is recalculated each time the feed is updated, making it possible to identify a drift before it affects overall performance.
Each signal is accompanied by its explanatory variables and its confidence interval, so that the final decision remains reasoned and verifiable.
Execution thresholds are defined by the user. The automation applies the rule without modifying its parameters, and each action remains viewable retrospectively.
The signals produced by Nourrétance are based on identifiable data sources: aggregated market flows, public macroeconomic indicators and volume data from the relevant stock exchanges. No proprietary, unverifiable sources are included in the calculation of the final score.
Each model is tested by back-simulation over distinct historical periods, then evaluated in walk-forward validation in order to limit overfitting to past data.
The review and documentation processes are inspired by practices used on institutional desks, with full traceability of model versions deployed in production.
The calculation frequency remains the same; it is the configured analysis horizon that determines the signal reading.
The trader configures a minimum confidence threshold for each asset class followed. When a signal exceeds this threshold, the alert is transmitted with its original latency and probability interval, allowing a quick decision without having to manually interpret dozens of technical indicators.
For a portfolio manager, the signal is aggregated over a weekly or monthly horizon. The tool identifies gaps between the target allocation and the actual allocation, and proposes an adjustment based on the evolution of the measured risk rather than on a one-off market reaction.
Public macroeconomic indicators are integrated into the same pipeline as market data, which makes it possible to situate a short-term signal within an underlying trend, useful for sectoral or geographic exposure decisions.
Access to the platform includes initial configuration of risk thresholds and a review of the validation methodology with an analyst.
