Surface Explorer

An important premise. QANTHOS is an analysis, validation and discovery tool, not a financial advisor. Surface Explorer doesn’t prescribe a setup: it identifies statistical patterns (plateaus vs peaks) in the surface you provide. The final choice of parameters, and any judgement on their adequacy, remain yours.

1. Where it starts: MultiCharts’s Optimization Report

Surface Explorer reads an Optimization Report exported from MultiCharts: the table listing every parameter combination tested during an optimization, with a column per parameter and a column for each performance metric (Net Profit, Profit Factor, % Profitable, Max Intraday Drawdown, and so on).

Two ways to load it, from the top toolbar:

• Paste from clipboard — you copy the table directly from MultiCharts and paste it here.
• Load file — open a file exported to disk.

The parser automatically recognizes which columns are parameters (those whose name ends with the strategy name in parentheses, e.g. iLongStartTime (MR_TF_Engine)) and which are metrics (the remaining numeric columns). If discarded rows or non-numeric values come up during parsing, the warnings are listed and shown explicitly — there’s never a silent data drop.

2. The key concept: plateau vs isolated peak

Surface Explorer’s core is distinguishing two situations that, looking only at “the table’s best value”, look identical but aren’t at all:

• The isolated peak (“naive max”) — the single parameter combination with the metric’s absolute highest value. It’s the number that would jump out sorting the table by descending Net Profit. But if the immediately neighbouring combinations give much worse results, that number might be a statistical artefact: a case where the optimizer found a point that works “by chance” on that specific history.
• The robust plateau — a wide zone of the surface where values are consistently good, not just at the best point but around it too. If you slightly shift parameters within a plateau, performance stays similar. This tolerance to small changes is precisely what you’d expect from a real edge, which is unlikely to depend on a parameter value chiselled to the tenth decimal.

Surface Explorer computes both and shows them together on the same surface, so the comparison — “does the naive max coincide with a plateau, or is it an isolated island?” — is immediate.

3. How plateau detection works

For those who want to understand what’s under the hood (details are always available as tooltips on the plateau table’s columns, in the UI):

1. For each grid cell, QANTHOS computes the neighbours’ average in a 3×3 window (called NeighborMean) and their coefficient of variation (CV = standard deviation / mean): how high, on average, the values around that cell are, and how dispersed they are.
2. A threshold is set (by default the top 20% of NeighborMean): only cells above threshold are candidates for a plateau.
3. Adjacent candidate cells are merged into a single region (connected-components analysis).
4. Each plateau gets a score combining average height, extent (more cells = higher score) and stability (lower CV = higher score).
5. Based on size and CV, each plateau receives a synthetic verdict:

The exact numeric thresholds (minimum size, CV) are the ones reported in the plateau table’s tooltips in the UI — here we only cover the logic, so as not to duplicate a detail that could be refined in future versions.

With a single optimized parameter, the logic is the same but in one dimension: instead of 2D regions, plateaus are contiguous segments along the parameter’s axis, shown as a scatter plot with coloured bands.

4. Selectors and display modes

The left panel governs what you see:

• Parameter X / Parameter Y — the heatmap’s two parameters. With only one parameter in the dataset, Surface Explorer automatically switches to a 1D scatter.
• Metric — the performance metric to represent (Net Profit, Profit Factor, etc.).
• Aggregation — when several table rows share the same (X, Y) combination — e.g. a third parameter varied while X and Y stay fixed — this setting decides how to combine them into a single cell value (by default the median, more robust to outliers than the mean).

Three mutually exclusive display modes:

On the 2D heatmap, a count overlay (toggleable separately) shows how many table rows were aggregated into each cell — useful for understanding whether a “good” cell rests on a single observation or on many. An interactive tooltip on hover reports the exact parameter and metric values under the cursor.

5. Multi-metric mode: the anti-overfitting intersection

A zone might look like a solid plateau on Net Profit, but be fragile on Max Drawdown or % Profitable. If a parameter “works” only on one metric and not on the others, it’s more likely exploiting a statistical artefact of that specific metric rather than a real edge of the strategy.

Multi-metric mode addresses exactly this: you select up to 4 metrics (for each you can indicate whether it should be maximized or minimized — useful for metrics where “less is better”, like the number of consecutive losses), and Surface Explorer draws:

• a heatmap for each metric, with the top-X% cells outlined and a coloured star on the centre of that metric’s best plateau;
• an intersection panel: the cells that are in the top-X% on all metrics simultaneously. If the intersection is empty, there’s no zone satisfying all criteria at once — information as valuable as a zone found.

In multi-metric mode the individual heatmaps’ colour represents the cell’s percentile rank within its own metric (not the absolute value): this allows comparing metrics with very different scales (dollars, percentages, factors) by using the full colour range for each. A readability choice, also stated below the chart.

6. The plateau table

Below the chart, a table lists detected plateaus (up to the configured maximum), with an extra Naive max row for direct comparison:

Every column header has a tooltip with the full explanation, viewable on hover.

7. Exporting a result

The Export button saves the current result as JSON, to bring into documentation or compare in the future:

• in single mode, it exports the plateau selected in the table (or P1 by default, or the naive max if there are no plateaus);
• in multi-metric mode, it exports the current intersection with the thresholds used for each metric.

The file always includes the context (parameters, metric, aggregation, top-X% threshold) needed to interpret the result later.

8. Common issues and notes

“No data loaded”. You need to paste or load a valid Optimization Report from the toolbar first.

The Parameter Y combo is disabled. The dataset has only one parameter: Surface Explorer automatically switches to the 1D scatter, where the single parameter goes on the X axis.

The multi-metric intersection is empty. No cell satisfies the top-X% on all selected metrics simultaneously — an informative result, not an error: try raising the top-X% threshold or reducing the number of metrics required in AND.

Warnings appear after loading. The parser discarded rows or non-numeric values; details are in the message. Check the file exported from MultiCharts if the warning count is high.

The naive max and the best plateau don’t coincide. A normal, and actually instructive, situation: it means the absolute best point isn’t at the centre of a stable zone. Worth checking carefully whether the setup you had in mind was close to the naive max or to the plateau.

9. An example workflow

A typical path, not a prescription.

1. Run the optimization in MultiCharts and copy (or export) the Optimization Report.
2. In Surface Explorer, paste from clipboard or load the file.
3. Select the two parameters and the main metric; leave aggregation on median as a starting point.
4. Look at the heatmap: check whether the naive max falls inside a robust plateau or is isolated.
5. If you care about more than one metric, switch to multi-metric mode and check the intersection.
6. Use the plateau table to read the exact centre (X, Y) of the most solid zone.
7. Export the result if you want to keep it or compare it with other optimizations.

A robust plateau isn’t a guarantee of future profit: it’s an indication that the strategy’s behaviour doesn’t fragilely depend on the parameter’s exact value. It remains an analysis of historical data, though — the strictest validation comes from out-of-sample behaviour.

10. To learn more

• Bias Explorer — another Developer Assistant diagnostic analysis, on recurring time patterns.
• Strategies Tab — where the strategies you then optimize and check here come from.