The RQA tab

One recurrence plot per measure, with the windowed metrics beneath it and the playhead across both. This page is about what the tab draws; what the numbers mean and how they were computed is on the analysis page.

• Tab id rqa, label RQA Plots, order 20, drawn into rqaContainer.
• Gate: include_RQA must be an array with at least one entry. A bare true does not switch this tab on.
• Reads assets/rqa/{videoID}_rqa_data.json, container key rqa_data, one entry per data type.

What you see per measure

One block per key in rqa_data, each a single figure with four parts sharing one time axis:

part	what it is
the square	the recurrence matrix — a black cell wherever the measure was in a state it had been in before, white elsewhere. Both axes are time
top strip	the raw measure, in its colour from the time series tab
left strip	the same measure again, rotated, so a horizontal band on the square lines up with the value that produced it
four strips below	the windowed metrics, top to bottom: RR, DET, LAM, L_MAX. L_MAX is in seconds; the other three are unitless fractions

The heading above each square carries that measure's achieved recurrence rate as a percentage and the threshold that produced it, both to fixed precision.

The metric strips are tied to the square's x-axis, so the four of them and the top marginal are always aligned with the recurrence plot — you can read a spike in DET straight down onto the structure that caused it.

The square really is square. The figure computes its own pixel geometry rather than letting Plotly stretch it: the plotting width is 85 % of the space available after margins, floored at 300 px, and the square itself is 80 % of that again. The total height is derived from the result. A stretched recurrence plot misrepresents diagonal structure, which is the thing the measures are counting.

The cost is that the figure is drawn at an explicit pixel size and does not reflow. Because the width and height are both fixed in the layout, the host's resize pass cannot recompute it either — the figure only takes a new size when something redraws it outright, which means a playhead move, a theme change or a recording change. Resizing the window alone leaves it as it was.

The playhead

Clicking any subplot except the rotated left strip seeks to that time. The left strip is excluded because its x-axis is value, not time — a click there has no time to seek to.

A selected time draws four shapes, which is more than the other tabs because a recurrence plot is symmetric in time:

• a vertical band across the square, the top strip and every metric strip,
• a line at the playhead itself,
• a horizontal band across the square, at the same times,
• a matching horizontal band on the left strip.

The band is playhead ± windowSize / 2, clamped at both ends to the first and last time in the drawn series, so it never runs off the edge of the plot. Window size comes from the Window Size control, defaulting to 5 seconds.

onTimeUpdate re-renders every figure from scratch, decoding the matrix again each time. With several measures this is the most expensive redraw in the dashboard. It is why the tab feels slower to scrub than the cross-wavelet tab, which moves shapes instead.

What it does not do

• It does not compute anything. Everything drawn is read from the payload. If a measure is listed in include_RQA but absent from the payload, the tab logs the mismatch to the browser console and draws the rest — it does not warn you on the page.
• It does not draw the full-resolution matrix. The payload ships a matrix reduced to at most 500 × 500 cells, by a density-preserving reduction — so the picture keeps both its structure and its recurrence density. The percentage in the heading is the full matrix's rate, not the drawn one's. The drawn matrix's own rate is computed and stored as visualization.reduction.rate_drawn, and is never displayed; it is close but not identical, and it is measured a slightly different way — it includes the line of identity, which the headline rate excludes. Do not read the heading as a description of the pixels. See the analysis page.
• It has no controls of its own. Everything it responds to — the recording, the playhead, the window — belongs to the page, not the tab.

When it cannot draw

The ones you are most likely to meet, each naming its own fix:

• No payload at all — include_RQA is set, so the analysis was expected. The message tells you to run python build_assets.py in the study folder.
• A payload from an older format — the host's version check refuses it and says so, rather than drawing something subtly wrong. See contracts/analysis-output.md.
• A payload missing visualization.time, .data, .matrix_size or .matrix — that one measure's block shows the error and the others still draw.
• A payload whose rqa_data is empty — reported as empty or invalid, with the structure logged to the browser console.

Where it lives

packages/dims-tabs/rqa.js, which adds loadRQAData, displayRQAPlots, createRQAPlot and updateRQAHighlights to the host. The figure itself is built by _renderRecurrenceFigure, which lives on the host and is shared with the cross-RQA tab.