Lturner applies predictive models to structured and unstructured market data, producing recommendations that are logged, dated, and reviewable. No narrative is attached to a result after the fact.
The process is sequential and auditable. Each stage produces a record that feeds into the next, so a given output can always be traced back to its inputs.
Market feeds, order book depth, and macroeconomic releases are normalised into a common schema before any model sees them.
Statistically relevant variables are derived from raw data — volatility clustering, liquidity shifts, correlation drift — and weighted by recency.
An ensemble of predictive models scores probable outcomes across defined time horizons, each output tagged with a confidence interval.
Recommendations that fall outside predefined risk thresholds are flagged or discarded before reaching the dashboard.
The final recommendation, along with its supporting data trail, is written to the daily audit log — unedited.
Every output the engine generates is recorded on the day it is produced, independent of how the market subsequently behaves.
Each entry includes the input data snapshot, the model version used, and the resulting output, time-stamped and stored without retroactive adjustment.
Model output is scored against actual market movement over the same horizon, with results aggregated weekly to reveal drift or degradation early.
When volatility or correlation exceeds a configured threshold, the system issues an alert rather than silently adjusting its own parameters.
The same underlying engine supports two distinct audiences, each using its output for a different kind of decision.
Position sizing and rebalancing recommendations are generated from volatility-adjusted risk scores rather than fixed allocation rules, allowing exposure to shift as conditions change.
Applied to multi-asset portfolios where correlation between holdings needs to be reassessed more frequently than a quarterly review allows.
Business leads use the same data layer to monitor sector-level demand signals and pricing pressure, informing decisions outside of direct trading activity.
Suited to teams that need a structured view of market conditions before committing capital to a strategic initiative.
The system presents a ranked set of options with its reasoning attached. The final action remains with the person reviewing the dashboard.
Designed for users who want to reduce the cognitive load of monitoring multiple data streams, without removing themselves from the decision.
We do not claim a fixed rate of return. What follows is a description of how the models are validated and how often results are disclosed.
Each model version is back-tested against historical data it was not trained on, and the results of that test are retained alongside the model's deployment record.
Input data is sourced from licensed market feeds. Any gap, delay, or anomaly in a feed is flagged in the audit log rather than interpolated silently.
Performance summaries are issued daily. A consolidated review covering model accuracy and risk events is issued weekly.
The interface is deliberately plain. Figures are presented without annotation, so that a user's own judgement, rather than a dashboard's framing, determines how a result is interpreted.
We assume our users will check the numbers against their own records. The audit log exists to make that checking straightforward.
Documentation covering model validation, data sources, and historical reporting is available on request. Dashboard access is provisioned after a short onboarding step.