Provable Metrics

Every figure on the board can be verified. Page views add up, referrer counts add up, and resolution class counts add up to the same total. The numbers check each other.

What makes a metric provable

A metric is provable when you can verify it from the underlying data without relying on an algorithm you cannot see or test. The simplest provable metric is a count: if the board says 300 page views were recorded for /about on a given day, that means 300 events arrived for that path on that date. You cannot look into the data store and count the events yourself as a practical matter, but the figure is a direct count with no statistical adjustment, sampling, or de-duplication applied.

The property that makes the counts meaningful is additivity. The total page views for a period should equal the sum of page views for each individual page in that period. The total should also equal the sum of the daily page-view counts for each day in that period. If those sums do not agree, something is wrong. Because the figures are plain counts, you can verify the arithmetic yourself just by adding the numbers shown on the board.

This stands in contrast to metrics that involve statistical estimation. A unique visitor count derived from sampling a subset of traffic and extrapolating to the full population is not directly verifiable from the data shown. A de-duplicated session count that uses probabilistic matching across cookies is not independently checkable. The figures on this board are not estimated or de-duplicated. They are direct counts, and they add up.

Checking the arithmetic on the board

To verify the page-view total for any date range, add the page-view counts for each path shown in the page list. The sum should match the total shown at the top of the board for that range. If the board is showing 10 paths plus a remainder grouped as "other," the sum of those 10 plus the "other" figure should equal the total.

To verify from the time dimension, the sum of the daily bars in the visits-by-day chart should also equal the same total. A total of 2,400 page views for a 30-day range means the daily bars average 80 views per day, and their individual values should add up to 2,400. Checking this across both dimensions — by page and by day — confirms the counts are internally consistent.

The referrer counts and the resolution class counts are not expected to add up to the page-view total. Referrers and resolution classes are separate dimensions. Each page-view event contributes one count to its resolution class bucket and one count to its referrer bucket, but those buckets are separate tallies. The referrer total does add up to the page-view total because each event has exactly one referrer (which may be "direct"). The resolution class total also adds up to the page-view total for the same reason.

Why provability matters for a site owner

Provability builds trust in the numbers. If you cannot understand how a figure was derived, you cannot know whether to trust it. When the board shows a spike in traffic on a specific day, you want to know: is this real, or is it an artifact of a measurement algorithm? With plain counts, you can check. Did the page-view total go up? Did specific paths drive the increase? Does the referrer list show a source that would explain it? Those three checks take thirty seconds and confirm or explain the spike.

Provability also makes it easier to explain the numbers to other people. If a client, employer, or collaborator asks how you know the site received a certain number of page views, the answer is straightforward: the tracking script fired that many times on that day, and each firing incremented a counter. There is no sampling methodology or de-duplication algorithm to explain or defend. The count is the count.

For operators who use the board to make content or business decisions — should we invest in more articles like this one, did this traffic campaign work, is the new page design drawing more visitors — provable numbers make those decisions cleaner. You are not reading a dashboard that might be giving you a statistical estimate. You are reading direct counts of actual events.

The limits of what counting can prove

Provability applies to what is counted, not to what is not counted. If the tracking script fails to fire on some pages — because it is missing from a template, blocked by an ad blocker, or loaded in a way that prevents execution — those page loads are not in the count. The count is accurate for the events that were received, but it does not prove that all page loads generated events.

The gap between page loads and recorded events is worth keeping in mind when reading traffic figures. Ad blockers and script blockers prevent the tracking script from running on some browsers. The proportion of your audience using such tools varies by site topic and audience demographics. For technical audiences, the blocked fraction may be meaningful. For general-interest content, it is typically a small percentage.

The provable quality of the counts means you can trust the figures for what they represent: events successfully received and recorded by the server. What you cannot infer is a precise total for all page loads that occurred on your site, including untracked ones. This is a genuine limitation, and it is honest to name it. The counts are complete and accurate for the events they contain; they are not a guarantee of total coverage.

Troubleshooting

If the page-view total on the board does not match the sum of the individual page rows, check whether the board is applying a filter or showing only the top N pages with an "other" group. The "other" group contains paths that fell below the display threshold. Adding the visible paths plus the "other" figure should produce the total. If the arithmetic still does not work, use the feedback link to report it.

If the sum of daily bars in the visits-by-day view does not match the total shown for the same date range, confirm that the date range selected is exactly the same for both views. A one-day offset in the range will cause a mismatch equal to one day's worth of traffic. Adjust the range so both views cover exactly the same start and end date.

If you are seeing unexpectedly high counts and suspect bot traffic or script errors are inflating the figures, check the referrer breakdown for unusual sources. Automated crawlers often show up in the referrer list with distinctive domain patterns. Legitimate crawler traffic from search engine indexers is normal and typically represents a small fraction of total counts; unusually high counts from unknown referrers may indicate a spike in automated requests.