A check is a week, not a day

Rankgap checks each keyword you track once a week, and it counts that week from the keyword’s own last check rather than from a calendar Monday. Add one keyword today and another on Wednesday, and the two keep being checked on different days for as long as you track them.

Why your whole list on two dates is the wrong comparison

So the tracker does not take a visibility change by comparing your whole list on one date with your whole list on another. On any given date only part of your list was checked, so “last Tuesday against this Tuesday” is two different subsets of your keywords, and the difference between them reads as a change in your rankings when it is a change in which keywords were asked about.

The visibility figure groups checks into seven-day rounds instead. Round zero is the seven days ending at your newest check; the round before it is the seven days before that, and so on back through the window a screen reads. Inside a round each keyword counts once, at its newest check in that round, so a keyword checked twice in one week cannot weigh twice.

Two dates do get compared, and the comparison that is safe is the one scoped to a single keyword. A mover — the rows under “Biggest movers”, and the movement line in the weekly summary — is one keyword’s newest check against that same keyword’s previous one. There is no subset problem there, because both readings are of the same keyword. What is not done is the same comparison across a list whose members were checked on different days.

A week with nothing in it is a gap, not a zero

A round that holds no checks at all is left out rather than drawn at zero. A week nobody measured is not a week your visibility collapsed, and drawing it as one would be the chart making a claim the data does not support. On a line that shows as a break; in a figure it shows as the change simply not being there that week.

What a change is taken over

A change is taken over the keywords both weeks checked. A keyword you added this week moves the figure, because the figure is over this week’s keywords — but it cannot move the change, because there is nothing from last week to compare it against. Where those two sets differ, the screen says how many keywords the change is actually over.

And until a second weekly round has landed, there is no figure at all: the screen shows a dash and says what turns it on. One round is a reading, not a trend.

A tracker that stopped

If the checks stop arriving for long enough that the newest round is plainly old, the last change is still printed, with the date it belongs to, and without the green or red it would otherwise carry. A colour says “this happened”; a month-old arrow in green says it happened now.

None of this is a limitation we are working around. A weekly instrument whose rounds mean the same thing every week is the thing that makes a change worth printing, and the price of that is admitting the weeks it has nothing to say about.