See the shape of a research problem.
Quantitative research is easier to reason about when uncertainty, disagreement, parameter behavior, and distributions are visible. This lab turns those ideas into interactive diagrams rather than decorative charts.
One point is simple. A population has structure.
This first scene is deliberately not a performance leaderboard. It demonstrates how many candidate research objects can occupy several analytical dimensions at once. The axes below are synthetic variables so the visual can teach spatial reasoning without implying a market result.
A single path hides the range of possible paths.
A Monte Carlo view is useful because one sequence can look unusually smooth or unusually poor by chance. This illustration generates multiple synthetic zero-drift paths to make path dispersion visible. It is not a forecast and does not use market data.
Synthetic Monte Carlo field
64 paths · 90 synthetic steps · deterministic seed until regenerated
A broad neighborhood tells a different story than one sharp peak.
This synthetic terrain compares response shapes across two normalized parameters. Switch between a broad plateau, a narrow spike, and a ridge. Height is an illustrative response score, not strategy performance.
Synthetic robustness surface
21 × 21 parameter grid · deterministic functions · no market data
Diversification is a network problem, not a list of names.
This deterministic synthetic network groups eighteen series into three families. Lines appear only for strong illustrative relationships. Cycle through nodes to see how one series can sit inside a dense cluster even when the whole graph looks diversified.
Synthetic correlation network
18 series · 3 synthetic families · edges shown above |relationship| 0.62
Averages can look similar while tails behave very differently.
Switch among three deterministic synthetic distributions, then move the threshold to inspect the share of observations below it. The exercise is about shape, skew, and tail mass—not forecasting returns.
Synthetic distribution explorer
4,000 observations · deterministic seed · standardized illustrative outcomes
A regime is a coordinate system, not a hindsight label.
This synthetic cube places observations on three normalized dimensions: directional tendency, volatility, and liquidity stress. Rotate the cube, filter stress conditions, and inspect points to see how a market-state label can compress several variables into one word.
Synthetic regime cube
96 deterministic observations · 3 normalized axes · illustrative thresholds only
Move the boundary through time instead of trusting one split.
This synthetic landscape shows seven ordered train/evaluate folds. Switch between rolling and anchored training windows, then inspect each fold. The geometry explains data partitioning only; it does not display strategy performance.
Walk-forward window landscape
100 synthetic time units · 7 folds · ordered partitions · no market data
A holdout stops being independent when it starts steering selection.
This synthetic diagram contrasts a preserved holdout with repeated peeking. In the clean case, candidate selection happens before the boundary and the chosen candidate crosses once for evaluation. The repeated-peek mode adds illustrative feedback loops from holdout information back into selection.
In-sample / out-of-sample boundary
36 synthetic candidates · one evaluation boundary · no measured strategy results
Similar endpoints can hide very different time spent underwater.
Drawdown is a path property. These deterministic synthetic index paths show a deep shock, a long shallow valley, and repeated valleys. The shaded distance from each running peak makes depth and recovery visible without treating a single maximum-drawdown number as the whole story.
Drawdown valley explorer
120 synthetic path steps · running-peak reference · no market or account data
Search more candidates and the best-looking noise gets more extreme.
This synthetic independent-null demonstration draws candidate scores from a standardized Normal distribution with no signal. Increase the family size to see how selection alone can produce a more impressive maximum. The approximation is educational and depends on the stated independence/Normal assumptions.
Selection / multiple-testing funnel
Independent synthetic Normal null · deterministic seed per family size · no strategy results
Where to go next
Use Research Lab to explore the shape of a problem, then move to Methods for assumptions, Learn for deeper study, or Docs for product reference.