Exploring data with the Para-Sieve

The Para-Sieve is the interactive chart at the heart of the viewer. It is a parallel-coordinates plot: every column in your dataset becomes one axis, the axes are stacked across the plot, and every row in the dataset is drawn as a single line that threads through all of them — passing each axis at the height of that row's value for that column.

One line is one candidate building; one axis is one variable. Because every row crosses every axis at once, you can see relationships between many variables in a single picture — which inputs tend to drive a low energy result, where the trade-offs lie, and which rows are outliers. The name is the idea: you use the axes as a sieve, narrowing many rows down to the few that matter.

You will meet it in two places. The sample page fills it with randomly generated data so you can learn the controls risk-free; nothing there is real, and its bookmarks are cleared when you leave. The viewer shows a real dataset you have uploaded to a project, and remembers your bookmarks for that dataset between visits. The controls are identical in both.

Before touching anything, read what is on screen:

• Each axis runs across the plot with its own scale. Its name — and unit, where it has one — sits at the end of the axis. The smallest value is at one end, the largest at the other.
• Each line is one row, drawn from the first axis to the last. Where the line crosses an axis is that row's value for that column.
• Lines that run roughly parallel between two axes are positively related; lines that cross over in an X between two axes are inversely related. Spotting those patterns is the whole point of the chart.

That is the static picture. Everything else on this page is about paring it down to the rows you care about, and reading individual rows out of it.

Filtering is the core move. Every axis carries a range slider with two handles — a low handle and a high handle. The band between them is the range that passes; rows whose value on that axis falls inside the band are kept, and the rest are filtered out.

1. Drag the low handle in from its end to raise the minimum, or drag the high handle in to lower the maximum. The plot updates live as soon as you release the handle.
2. A label appears for the handle as you move it, showing the exact value (and unit) it currently sits at, so you can dial in a precise cut-off.
3. Combine filters across several axes. Each new constraint is applied on top of the others — only rows that satisfy all of your active filters survive. This is how you sieve a crowd of rows down to a handful.

Inverting a filter

Sometimes you want everything except a middle band — the extremes rather than the centre. Each axis has an invert button (the chevrons at the end of the axis) that flips the filter:

Normal (chevrons pointing outward)

Keep rows inside the band between the two handles.

Inverted (chevrons pointing inward)

Keep rows outside the band — the two tails instead of the middle.

You can also right-click the slider track to toggle inversion without aiming for the button. The shaded part of the track always shows which side currently passes, so the picture matches the rule at a glance.

As you filter, the results panel below the plot keeps a running tally of which rows survive and lists them in a table — one row per surviving line, every axis value spelled out in full. It is how you read the exact numbers behind the lines you have isolated.

The panel guides you through the sieve with a short status message:

Apply a filter to see results here.

You have not narrowed anything yet — start dragging a handle.

A row count, e.g. 47 rows

Few enough rows pass that they can all be listed. The table shows them. This is the goal — you have sieved the data down to a readable set.

Too many rows still pass the filters…

More than the panel can list (about 100 rows) still pass. Tighten a filter or two to bring the count down into the listable range.

No rows match the current filters.

Your filters are mutually exclusive — nothing satisfies all of them at once. Loosen one and the rows reappear.

By default the lines share a neutral colour. Pick a colouring axis and every line is instead tinted by its value on that one axis — low values at one end of the spectrum, high at the other — turning a single variable into a colour gradient laid across the whole plot. It is the fastest way to see how one result spreads through everything else.

1. Open the palette panel using the palette trigger on the left edge of the plot. A palette button appears beside each axis.
2. Click the palette button next to the axis you want to colour by. The lines recolour immediately, graded by that axis's value.
3. Choosing a different axis moves the colouring; only one axis colours the plot at a time.

The colouring axis does double duty: when it is set, the results panel sorts its rows by that axis's value too, and notes which axis it is ordered by in the header.

A wide dataset can crowd the plot with axes you are not currently interested in. Hiding an axis removes it from the chart — its filter still applies, it is just out of the way visually.

1. Open the toggles panel using the eye trigger on the right edge of the plot. An eye button appears beside each axis.
2. Click an axis's eye button to mark it hidden; click again to mark it shown. The button switches between an open and a crossed-out eye so you can see each axis's state at a glance.
3. Close the panel to apply your choices. The axes you marked hidden now drop out of the plot.

Why hiding is deferred until you close the panel

The toggles panel is a staged toggle system, and this is the key thing to understand about it: while the panel is open, every axis stays on screen — even the ones you have marked hidden — each showing whether it is currently set to hide or show. The axes only actually leave the plot once you close the panel away.

That is deliberate. Because a marked-hidden axis stays put while the panel is open, you can always find the one you just hid — it is right where it has always been, in its actual context — and unhide it in place rather than hunting for it. Hiding and un-hiding happen in the same spot, with nothing jumping around under your cursor.

It also keeps hiding, reordering, and colouring coherent: all three act on the same row of axes in the same panel context. Keeping every axis visible while you work means a drag-reorder still makes sense whether an axis is marked hidden or not — you are arranging the full set, then committing the visibility once you are done.

Relationships between two variables are clearest when their axes are adjacent — the crossing-over pattern only shows between neighbours. Reordering lets you park the two axes you are comparing side by side.

1. Find the grip handle to the left of an axis (the horizontal grip icon).
2. Press and drag it up or down. The axis follows your cursor and the others slide aside to make room — the new order takes effect live, as you drag.
3. Release to drop the axis into its new slot. The dropped axis glides into place.

To read a single row without filtering down to it, hover over its line. A tooltip appears naming the row and listing its value on every axis — the full record behind that one line.

Hover is controlled by the crosshair button in the toolbar above the plot:

• When it is on, moving your cursor near a line highlights it and pops the tooltip. As you slide between lines the values update, and any that changed flash briefly so the difference catches your eye.
• When it is off (the button is dimmed), the plot ignores the cursor — handy on a dense chart where every move would otherwise trigger a tooltip.

How close the cursor has to be, and how boldly the line is highlighted, are both adjustable — see settings.

A bookmark pins a row so it stays highlighted no matter how you filter afterwards. It is how you set aside a promising candidate and keep referring back to it while you explore the rest.

1. Hover over the line you want to keep, then press B. The row is bookmarked, and a bookmarks panel appears listing it with all its values.
2. Press B again while hovering the same line to remove the bookmark, or use the bookmark button in the panel to remove one directly.
3. Hover a row in the bookmarks panel to light up its line in the plot, so you can place it among the rest at a glance.

The settings button (the gear in the toolbar) opens a menu of appearance and behaviour controls. None of them change your data — they tune how the plot is drawn and how it responds — and each slider has a reset button to return it to its default.

Interpolation

Segmentation

How finely each line is drawn. Higher values give smoother curves at a small cost to rendering; lower values are coarser but lighter.

Curvature Intensity

How much the lines bend between axes. At the low end lines are nearly straight; turning it up folds them into pronounced curves. Its effect is purely visual — pick whatever reads most clearly for your data.

Overlay

Space Between Axes

The gap between neighbouring axes, in pixels. More space spreads the plot out and makes crossing patterns easier to follow on a wide dataset.

Axis Labels Visibility

Whether axis names show Show always, only While Hovered, or Hide never — trade a tidy plot against always-on labels.

Draw Non-Passing Rows

With Show, rows that fail your filters stay on the plot as faint ghosts so you keep a sense of the full distribution. With Hide, only passing rows are drawn, for an uncluttered view of the survivors.

Hover

Hover Detection Radius

How close the cursor must come to a line before it is picked up. A larger radius makes lines easier to catch on a sparse plot; a smaller one is more precise when lines are packed together.

Hover Line Thickness

How boldly the hovered line is drawn, so the one you are inspecting stands out from the crowd.