Uptime.com’s Metrics Explorer tool allows for highly granular reporting, comparison, and historical tracking of check metric data such as response time. The tool includes a chart builder that lets you combine multiple checks and metrics on a single chart, split data by dimensions such as monitoring location or transaction step, overlay an earlier period for comparison, and apply data filters for deep analysis.
Table of Contents
- Using Metrics Explorer
- Example Playbook: Using Metrics Explorer During an Incident
- Example Use Case and Guide: Browsing Transaction Check Metrics Step by Step to Troubleshoot Problem Steps
- Limits and Tips
Using Metrics Explorer
Open the tool in two ways:
- From a Check Analysis page, click the Explore Metrics button in the top right:
- Open the tool directly at the URL https://app.uptime.com/metrics-explorer, which will create a blank chart
Viewing the Metrics Explorer Page
Elements of the page are as follows:
Chart name: Click the chart name to open the chart menu: create a new chart, rename or delete the current one, or open a saved chart. Includes a summary of the number of series and lines on the chart (e.g., 1 series | 3 lines on chart)
Export: see Exporting Charts
Save: click the Save icon to give the chart a name and save; click the drop-down to Save as new. Each chart must have a unique name
Controls: select the Time range, Interval, Compare, and the Chart Type
Chart: drag across the chart to zoom in on a period; once zoomed, click Reset zoom to view the full range
Series panel: what the chart draws: one row/card per series. Click the Arrow icon to expand or collapse the card. Click the Eye icon to temporarily hide the card’s chart visualization. Click More Actions to open the menu to Duplicate or Remove the card.
Each series has a letter (A, B, C …) that also leads its lines' names in the legend and the tooltip, so you always know which row a line belongs to.
Analysis toggles: lists Analysis options (Trend, Moving average, Baseline figures, etc.). When an Analysis option is toggled on, the Results are displayed below the toggles.
Exporting Charts
Export a chart using the button on the upper right of the page. Charts can be exported as Chart and analysis, rendering a PDF with a visualization of the chart and analysis, or as Data and analysis, producing a spreadsheet. The export is computed from the data again rather than copied from the screen, so it is a faithful record.
Adding and Setting Up a Series
A series is one question you ask of the data: these checks, this metric, aggregated this way, optionally split and filtered. A chart can hold up to 12 series. All checks in a series must be of the same check type; adding checks of mixed types results in a warning:
Click Add series at the bottom of the list to add one. The new series opens and scrolls into view.
A series must have at least one check selected to begin configuring its options:
- Checks: one or more checks. They must share at least one kind of data
- Data: the kind of data to read, such as http, transaction (whole transaction) or
- transaction_step (each step). The list of available data types depends on the check type
- Aggregation: how the values in each time bucket are combined: Average, Minimum, Maximum, Sum, Count, Median, or the 75th, 90th, 95th or 99th percentile
- Metrics: one or more measurements from the chosen data, such as Response time or Step
- Duration
- Split by: draw one line per value of a dimension. Choose per monitoring location or per step. Up to two dimensions can be selected
- Filter by: narrow the series to chosen values of a dimension without splitting by it
As options are chosen, the explorer automatically fills in what it can. Picking a check fills in Data and Metrics with sensible defaults, so a chart appears straight away instead of more empty dropdowns. Switching to a check of another type keeps the series valid. A Data or Metric value that still applies is kept; one that does not is replaced by the only option, or the default. If the explorer has to choose among several options, the field is outlined and marked Picked for these checks until you confirm or change it.
Each series card has a Duplicate button (a quick way to make a variant, such as the same check with p95 instead of average) and a Remove button.
Time Range, Interval, and Comparison
Define your chart’s time, display interval, and choose comparison overlays using these options.
Time Range: select a preset (last 30 minutes up to last 90 days) or a custom range. The chart's time axis always matches the range you picked.
Interval: the size of each time bucket. Options include:
- Auto
- 1 minute
- 15 minutes
- 1 hour
- 1 day
- 1 week
- 1 month
Auto picks one that suits the range. Choose a smaller interval to see short spikes, a larger one to see the shape of a long period. A request can hold up to 2,000 buckets per series, so very fine intervals need shorter ranges (30 days at 1 minute is too many; the explorer tells you so).
Compare:: overlays the same chart for an earlier period. Options include:
- Previous period
- Previous day
- Previous week
- Previous month
- Previous year
Use it to answer is this normal for a Tuesday morning?, or to check that a deploy, a config change or a fix moved the numbers the way you expected. Above, this week ran about half a second slower than last week until 27 October, when last week's own step up caught up.
The tooltip shows when each compared point was observed. Hover over a point on the chart to see the breakdown of data from each point in the monitoring history:
Chart Types
Pick the chart type from the drop-down at the top right. The first five plot values over time; the last three read the data differently.
Line: the default and the most precise. Displays one line per series and split value. Best for watching a metric over time, reading exact values from the tooltip, and comparing a few lines. All the analysis overlays (trend, moving average, anomaly band) draw on it.
Area: a line with the area beneath it filled. Good for one or two series where the size of
the value matters — overall load time, total bytes — and for dashboards that are
read from a distance.
Stacked area: series are stacked on top of each other, so the top edge is their sum. Use it when the parts add up to a whole: transaction steps adding up to the full run, or response time phases. You see both the total and which part is growing.
Bars: one bar per time bucket. Clearer than a line for coarse intervals (per day, per week) and for counts, such as how many checks ran or failed in each bucket.
Stacked bars: bars with the series stacked, so each bar shows the whole and its parts for that bucket. Handy for day-by-day composition — for example how much of each day's transaction time each step took.
Scatter (correlation): plots the first two series against each other: one point per time bucket where both have a value, the first series on the horizontal axis and the second on the vertical. Points along a rising line mean the two move together.
Use Scatter to test a hypothesis — is the total slow because step 8 is slow? Does response time rise with response size? Is one location's latency tied to another's? Put the two things you want to compare first in the series list.
Distribution: a histogram. Shows how often each value occurs over the range, ignoring when. Use it to see the shape of performance — where most runs land, how wide the spread is, and how long the slow tail is. Tails are what percentile SLOs are about; a narrow peak with a long tail tells a very different story from a wide, flat spread with the same average. Use a smaller interval (1 hour or less) so there are enough values to bin. In the below chart, most hourly runs fall between 9.5 s and 10.5 s, with a thin tail out to 12.5 s:
Seasonality heatmap: A weekday-by-hour grid coloured by the metric's average in each slot, in your account's time zone. Requires that the Seasonality analysis is turned on. Use it to find patterns tied to the clock: busy hours, nightly jobs, backups, cron- driven batch work, weekend maintenance.
In the below example, Monday at 08:00 stands out, with smaller hot spots at 08:00 on Tuesday and Saturday — it’s worth asking what runs then. Knowing these patterns also stops you from paging on something that happens every Monday morning.
Analysis
Turn analyses on with the toggles under the chart:
Overlays are drawn on the chart and the figures appear under Results, one card per line. Everything is computed on the server over the range on screen, for the visible series (comparison lines are not analyzed).
Trend
Fits a straight line through the data and reports how fast the metric is changing. Data points are displayed as follows:
- Trend: change per day (positive or negative)
- 95% CI: the range the true slope very likely lies in. A range that does not cross zero means
- the change is real, not noise
- P-value: How likely a slope like this occurs by chance alone; below 0.05 is significant
- Fit (R²): how well a straight line describes the data, from 0 to 1.
- Change over window: how much the line moved from start to end, in percent
For SREs: Trend is the early warning for slow degradation that never trips an alert — a dependency getting slower week by week, a database growing, a page gaining weight. It turns a general sentiment like “it feels slower lately into +42 ms per day, and +13% this month” into an action item that you can put in a ticket and use for capacity planning. After a fix, it confirms the line is flat again.
Moving average
Draws a smoothed line: the average of the last N points at each point (set Window (points), from 2 to 200).
For SREs: smoothing hides one-off spikes so you can see the level the service actually runs at. Above, a 24-point window over hourly data (one day) ignores the spikes and shows a real step up between 20 and 22 October — the kind of change a deploy or a configuration change causes. Choose the window to match what you want to ignore: about an hour's worth of points to ride out blips, or a day's worth to see the daily level.
Anomaly Band
Draws an expected range around the series and counts the points outside it.
- Sensitivity (σ): sets how wide the band is, in standard deviations (0.5 to 10). Lower catches more; higher only the extreme
-
Baseline: sets how "normal" is measured:
- Mean and deviation — the classic average ± spread
- Median and IQR — based on the middle half of the data, so the spikes you are looking for do not stretch the band themselves
- Outliers: the number and share of points outside the band; Latest (σ) is how far the most recent point is from normal. A latest value near 0 means the check is behaving normally right now
For SREs: the band answers the question “is this unusual?” with a number instead of a guess. Over a short range with a fine interval, the band hugs normal behaviour, and an incident stands out clearly (see the incident walkthrough). Latest (σ) is a quick check during and after an incident: is the service back inside its normal range?
Baseline figures
Adds the summary of the range to Results: Mean, Median, IQR (the spread of the middle half), 95th percentile and Samples (with the number of gaps). They are also shown whenever the anomaly band is on.
For SREs: these are the numbers for an SLO review or an incident report — median 9.83 s, p95 10.51 s over the last 30 days. Compare the median with the p95: a large gap means a long tail that averages hide. The sample count and gaps tell you whether the check ran the whole time or whether data is missing.
Seasonality
Profiles the metric by hour of day and by weekday, in your account's time zone, and fills the seasonality heatmap. Results show Seasonality — how much of the variation is explained by the hour of day, as a percentage — and the busiest hour.
For SREs: a high seasonality figure means the time of day matters: alert thresholds and expectations should differ by hour, and capacity has to cover the peak, not the average. A low figure means the variation is not tied to the clock, so look elsewhere for causes.
Show downtime
Shades the periods when the check was down behind the chart. Available when the chart covers a single check.
For SREs: put cause and effect side by side — did response time climb before the outage (a warning sign you could alert on), or only during it? Did it recover fully afterwards?
Example Playbook: Using Metrics Explorer During an Incident
Following is a playbook, using a real spike on an internal Uptime.com transaction check on 27 October.
1. Open the explorer from the check. On the check’s analysis page, click Explore Metrics. You get the total and every step on one chart, over the range you were already looking at.
2. Is it unusual? Pick a short range (Last 24 Hours or Last 7 Days), set the interval to 15 minutes, and turn on Anomaly band with the Median and IQR baseline. Normal behaviour sits inside the band; the incident is the point far outside it — here a jump from about 10.4 s to 17 s.
3. Which step? Turn the anomaly band off, hide series A (the total) and look at series B, split by step. One line moves; the others are flat. Here, it is step 8:
4. When exactly? Drag across the spike to zoom in. Step 8 went from about 3.6 s to 10.5 s around 09:30 on 27 October, with a second, smaller spike around 13:45:
5. Where? Add Monitoring Location to Split by (or open a chart that already splits by location). If one location spikes and the others do not, the problem is likely regional — network, a CDN edge, a regional dependency — rather than the application itself. Here only Frankfurt (deu-fra) spiked:
6. Is it new? Set Compare to Previous week or Previous day. If the same thing happened at the same time before, it is probably a scheduled job, not a new fault — the seasonality heatmap answers the same question over a longer range.
7. Is it over? Watch Latest (σ) in the anomaly figures return to near zero, and the trend over the last hours go flat.
8. Write it up. Save the chart with a descriptive name, and Export it as a PDF (chart plus analysis figures) or Excel (data plus figures) for the incident report or the post-mortem.
Beyond Incidents
- Regressions and capacity. A monthly Trend on key checks shows slow degradation before users notice. Per-step trends show which page or API in a journey is getting slower.
- SLOs and tails. Use the 95th or 99th percentile aggregation, the Distribution chart and the baseline percentiles to see the tail your users actually feel.
- Change verification. After a deploy, compare with the previous day or week, and check that the moving average stepped down rather than up.
- Noise reduction. Seasonality tells you which spikes are routine, so alerts can be tuned around them instead of paging every Monday at 08:00.
- Dashboards. Save the charts you rely on and add them to a dashboard as widgets.
Example Use Case and Guide: Browsing Transaction Check Metrics Step by Step to Troubleshoot Problem Steps
Transaction checks record two kinds of data:
- transaction: the whole run: Total time to complete transaction
- transaction_step: each step separately. Step duration, with a Step Number dimension
When you open the explorer from a transaction check, it builds both for you: series A is the total transaction time and series B is the step duration split by step number, one line per step.
Following is a step-by-step guide to browsing and analyzing transaction checks step by step:
Narrow to the steps that matter. In series B, open the Step Number value filter and pick the steps you care about. Here steps 2, 4 and 8 are kept next to the total:
The picture says a lot at a glance: over the month the total rose from about 9.2 s to 10.4 s, and step 8 alone went from about 2.8 s to 3.6 s — most of the growth lives in one step.
See how the total is made up. Hide series A and switch to Stacked area or Stacked bars: the steps stack up to the full transaction, so you see each step's share and which one is growing.
Confirm the culprit. Filter B to a single step and choose Scatter: if the points line up, that step is what moves the total.
Split by location as well. Add Monitoring Location to Split by to see whether a step is slow everywhere or from one place only.
Per-step data is recorded by the newer monitoring locations. A transaction check that runs only from older locations has whole-transaction times but no step breakdown.
Limits and Tips
The following limits apply to the maximum viewable data in Metrics Explorer charts:
- Up to 12 series per chart, up to 50 checks per series and 100 in total, and up to two split-by dimensions per series.
- Up to 2,000 time buckets per series and 100,000 data points per chart. If you hit a limit, the explorer says which one; widen the interval or shorten the Time Range.
- One series split by location or step can draw many lines; the summary under the chart name shows how many (11 lines on chart). Use value filters to keep the chart readable.
- Very long ranges of per-step data are the slowest thing to query — start with the last 7 or 30 days.
- "No data for this check in this range" usually means the check was paused or did not run in that range; Show last 30 days widens the range in one click.
- The explorer and the check's own page may draw slightly different curves for the same range, because they can pick different intervals.
Comments
0 comments