Contour | Zyeon UI

ready — 2,000 sessions, two separate modes; filled bands, and the sample hidden because 2,000 dots is more ink than information

Session quality

Last 7 days · 2,000 sessions

1,980 sessions · 5 contours · 72 × 40 grid · 20 not drawn

Smoothing1.00×

90% · 2 regions
70% · 2 regions
50% · 2 regions
30%
10%

Each ring is the smallest area holding that share of the sample, so the labels are the sample itself and not a density nobody can picture. Kernel 0.27 s by 5.4% at 1.00×. 20 sessions not drawn: 20 outside the fixed domain.

Contours of Session quality, outermost first. A contour's share is measured back off the grid, so it is what the ring encloses rather than what was asked for.
Contour	Share of the sample	Separate regions	Share of the plotted area
1	90%	2	34%
2	70%	2	17%
3	50%	2	9%
4	30%	1	4%
5	10%	1	1%

ready — lines only over the raw scatter: 900 conversations on a curved ridge, both domains fitted

Session length vs messages sent

900 conversations · both domains fitted

900 conversations · 5 contours · 72 × 40 grid

Smoothing1.00×

90%
70%
50%
30%
10%

Each ring is the smallest area holding that share of the sample, so the labels are the sample itself and not a density nobody can picture. Kernel 1.3 min by 1.6 at 1.00×.

Contours of Session length vs messages sent, outermost first. A contour's share is measured back off the grid, so it is what the ring encloses rather than what was asked for.
Contour	Share of the sample	Separate regions	Share of the plotted area
1	90%	1	26%
2	70%	1	13%
3	50%	1	8%
4	30%	1	4%
5	10%	1	1%

ready — under the sample floor: 14 readings, so nothing is smoothed and the points are drawn as they are

First week on the new tier

Pilot cohort · 14 sessions

14 sessions · below the 30-session floor · no estimate

Too few readings to smooth. 14 of the 30 a density estimate needs, so the sample is drawn as it is. A kernel over this many readings would draw a confident hill around whichever gap the sampling happened to leave.

ready — degenerate sample: 120 probes on one exact position, so Silverman's bandwidth is zero and the grid's resolution floor takes over; with so few occupied nodes the rings enclose 91/75/56/34/18% rather than the 90/70/50/30/10% asked for, and say so

Calibration sweep

Synthetic rig · 120 probes on one exact position

120 probes · 5 contours · 72 × 40 grid

Smoothing1.00×

91%
72%
56%
31%
16%

Each ring is the smallest area holding that share of the sample, so the labels are the sample itself and not a density nobody can picture. Kernel 0.06 s by 1.9% at 1.00×. That is the finest this grid can carry — raise the resolution to smooth less.

Contours of Calibration sweep, outermost first. A contour's share is measured back off the grid, so it is what the ring encloses rather than what was asked for.
Contour	Share of the sample	Separate regions	Share of the plotted area
1	91%	1	0%
2	72%	1	0%
3	56%	1	0%
4	31%	1	0%
5	16%	0	0%

ready — the same 2,000 sessions on a coarse 32-node grid with 3 levels: the same two modes, drawn with visibly straighter, more polygonal rings

Session quality

Last 7 days · 2,000 sessions

1,980 sessions · 3 contours · 32 × 18 grid · 20 not drawn

Smoothing1.00×

90% · 2 regions
50% · 2 regions
11%

Each ring is the smallest area holding that share of the sample, so the labels are the sample itself and not a density nobody can picture. Kernel 0.27 s by 5.4% at 1.00×. 20 sessions not drawn: 20 outside the fixed domain.

Contours of Session quality, outermost first. A contour's share is measured back off the grid, so it is what the ring encloses rather than what was asked for.
Contour	Share of the sample	Separate regions	Share of the plotted area
1	90%	2	36%
2	50%	2	9%
3	11%	1	1%

loading — the frame and a pulsing set of rings, so nothing shifts when the estimate arrives

Session quality

Last 7 days

empty — no readings to estimate

Session quality

Last 7 days

Nothing to estimate yet

A density estimate needs readings with a position on both axes.

error — click Try again

Session quality

Last 7 days

Couldn't load this density plot

The data source didn't respond.