:hedgehog: PostHog is the leading platform for building self-driving products. Our developer tools – AI observability, analytics, session replay, flags, experiments, error tracking, logs, and more – capture all the context agents need to diagnose problems, uncover opportunities, and ship fixes. Steer it all from Slack, web, desktop, or the MCP.
Python ab-testingai-analyticsanalyticscdpdata-warehouseexperimentsfeature-flagsjavascriptproduct-analyticspythonreactsession-replaysurveystypescriptweb-analytics
import { useMemo } from 'react'
import { BoxPlot, TooltipSurface, useChartTheme } from '@posthog/quill-charts'
import type { BoxPlotDatum, BoxPlotSeries, BoxPlotTooltipContext } from '@posthog/quill-charts'
import { cn } from 'lib/utils/css-classes'
export interface BoxPlotBucket {
/** x-axis and tooltip label for the bucket (e.g. its date). Must be unique across buckets. */
label: string
/** Samples in the bucket; 0 renders an empty slot. */
count: number
minSeconds: number | null
p05Seconds: number | null
p25Seconds: number | null
p50Seconds: number | null
meanSeconds: number | null
p75Seconds: number | null
p95Seconds: number | null
maxSeconds: number | null
}
export interface LeadTimeBoxPlotProps {
/** Series identity: chart key and the label shown in the tooltip legend. */
seriesKey: string
seriesLabel: string
/** One entry per bucket, oldest first. */
buckets: BoxPlotBucket[]
/** Value formatter for the tooltip rows (seconds in, short label out). */
formatSeconds: (seconds: number) => string
/** Draw the whiskers at p5/p95 instead of min/max, so one extreme PR can't flatten the boxes. */
excludeOutliers?: boolean
/** List the buckets down the side with the time axis along the bottom. Suits a few buckets. */
horizontal?: boolean
/** Put the time axis on a log scale, so minutes and days both stay readable on one axis. */
logScale?: boolean
dataAttr: string
className?: string
}
/** Per-bucket sample counts, carried into the tooltip through the series meta. */
interface BucketMeta {
counts: number[]
}
/** The six numbers a box draws, for the checks that have to cover all of them. */
const BOX_STATS = ['min', 'p25', 'median', 'mean', 'p75', 'max'] as const
function toDatum(bucket: BoxPlotBucket, excludeOutliers: boolean): BoxPlotDatum | null {
if (
bucket.count === 0 ||
bucket.minSeconds == null ||
bucket.p25Seconds == null ||
bucket.p50Seconds == null ||
bucket.meanSeconds == null ||
bucket.p75Seconds == null ||
bucket.maxSeconds == null
) {
return null
}
// Older cached responses may lack p5/p95; fall back to the full range rather than dropping the bucket.
const lower = excludeOutliers ? (bucket.p05Seconds ?? bucket.minSeconds) : bucket.minSeconds
const upper = excludeOutliers ? (bucket.p95Seconds ?? bucket.maxSeconds) : bucket.maxSeconds
return {
min: lower,
p25: bucket.p25Seconds,
median: bucket.p50Seconds,
mean: bucket.meanSeconds,
p75: bucket.p75Seconds,
max: upper,
}
}
/**
* One box-and-whisker per bucket (quill BoxPlot): whisker min→max, box p25→p75, a median line
* and a mean dot, on a shared seconds scale. Empty buckets stay empty slots so a quiet stretch
* reads as "nothing deployed", not missing data. One lead-time stage per instance; the Deploys
* tab stacks three vertical ones so the stages compare bucket by bucket.
*/
export function LeadTimeBoxPlot({
seriesKey,
seriesLabel,
buckets,
formatSeconds,
excludeOutliers = false,
horizontal = false,
logScale = false,
dataAttr,
className,
}: LeadTimeBoxPlotProps): JSX.Element {
const theme = useChartTheme()
const labels = useMemo(() => buckets.map((bucket) => bucket.label), [buckets])
const series = useMemo<BoxPlotSeries<BucketMeta>[]>(
() => [
{
key: seriesKey,
label: seriesLabel,
data: buckets.map((bucket) => toDatum(bucket, excludeOutliers)),
meta: { counts: buckets.map((bucket) => bucket.count) },
},
],
[seriesKey, seriesLabel, buckets, excludeOutliers]
)
const hasZeroDuration = useMemo(
() => series[0].data.some((datum) => datum != null && BOX_STATS.some((stat) => datum[stat] <= 0)),
[series]
)
const config = useMemo(
() => ({
yTickFormatter: formatSeconds,
axisOrientation: horizontal ? ('horizontal' as const) : ('vertical' as const),
// A log axis has no zero: the scale clamps a zero-second stat onto the axis floor, where
// it would read as a real duration. Such a bucket falls back to the linear axis.
yScaleType: logScale && !hasZeroDuration ? ('log' as const) : ('linear' as const),
}),
[formatSeconds, horizontal, logScale, hasZeroDuration]
)
return (
// The chart's root is a `flex-1` child, so the sized wrapper must be a flex column —
// in a plain block parent the canvas measures 0px tall and paints nothing.
<div className={cn('flex h-48 flex-col', className)}>
<BoxPlot<BucketMeta>
series={series}
labels={labels}
theme={theme}
config={config}
dataAttr={dataAttr}
tooltip={(ctx) => (
<BucketTooltip
ctx={ctx}
dataAttr={dataAttr}
formatSeconds={formatSeconds}
excludeOutliers={excludeOutliers}
/>
)}
/>
</div>
)
}
function BucketTooltip({
ctx,
dataAttr,
formatSeconds,
excludeOutliers,
}: {
ctx: BoxPlotTooltipContext<BucketMeta>
dataAttr: string
formatSeconds: (seconds: number) => string
excludeOutliers: boolean
}): JSX.Element | null {
const entry = ctx.seriesData[0]
const datum = entry?.series.meta?.datums?.[ctx.dataIndex]
if (!datum) {
return null
}
const count = entry?.series.meta?.user?.counts?.[ctx.dataIndex] ?? 0
const rows: [string, number][] = [
[excludeOutliers ? '95th percentile' : 'Max', datum.max],
['75th percentile', datum.p75],
['Median', datum.median],
['Mean', datum.mean],
['25th percentile', datum.p25],
[excludeOutliers ? '5th percentile' : 'Min', datum.min],
]
return (
<TooltipSurface data-attr={`${dataAttr}-tooltip`}>
<div className="font-semibold">{ctx.label}</div>
<div className="mb-1 opacity-70">
{count} PR{count === 1 ? '' : 's'} deployed
</div>
<table className="border-collapse">
<tbody>
{rows.map(([label, value]) => (
<tr key={label}>
<td className="pr-3 opacity-70">{label}</td>
<td className="font-medium">{formatSeconds(value)}</td>
</tr>
))}
</tbody>
</table>
</TooltipSurface>
)
}