Build signals, compare backtests and investigate execution with the same q/k array language. l puts column operations, numerical work and compression-aware execution underneath the research loop, so the next question stays close to the data.
01 / The workload
A useful hypothesis rarely arrives in one query. It takes repeated passes over instruments, dates, features and strategy outputs. When every pass expands columns or builds another intermediate dataset, the cost of iteration becomes part of the research decision.
Transform price, volume and derived columns into candidate signals. Group observations by instrument or research period and examine how relationships change across the universe.
Build the array calculations behind a strategy simulation. Compare your strategy outputs across parameter choices, holding periods and transaction-cost assumptions.
Analyze prepared trade-and-benchmark datasets to understand the gap between a research price and an achieved execution. Break results down by venue, size or market condition.
Aggregate positions, calculate exposures and compare supplied scenario outputs. Move between a portfolio summary and the instruments contributing to it.
02 / From data to analysis
A typical signal investigation moves from a defined universe to a feature panel, then to comparisons that remain traceable to the observations.
Start with normalized observations and the identifiers, time ranges and reference fields your strategy requires.
Use qSQL filters and groups alongside vector arithmetic, correlations and reductions to construct a feature panel.
Analyze the panel and your backtest outputs by instrument and period, then return a focused result to your research tools.
/ Filter a research panel l>r:([]sym:`A`B`A`B;signal:1 3 -1 5) l>select avg signal by sym from r sym| signal ---|------- A | 0 B | 4 l>select from r where signal>0 sym signal ---------- A 1 B 3 B 5 l>exec sum signal from r 8j l>
03 / Why l
You write q/k. The runtime considers the physical representation of the data, choosing encoding-aware operations where supported and materializing values when needed.
04 / Your first evaluation
Choose a representative research period, a small instrument universe and one feature or backtest calculation you already trust.
Available as a research preview for local evaluation.
How physical layouts become part of execution.
Read the paper Integration guideCall l from Python, pass typed values and bring results back to your application.
Read the guide Integration guideBring flat columnar datasets into l through the native Parquet reader and writer.
Read the guideYour data. Your next question.