What the numbers mean

Every figure RoutingCat reports is counted, predicted or measured, and each kind answers a different question. This page explains them, so you can say what a figure shows and how sure it is. Read and explain results applies it to the command’s output, and Read the dashboard to the dashboard’s.

Three kinds of number

Kind	What it says	Examples	How sure
Counted	What happened	Visits, conversions, rates, revenue, the share of its slot a variant was served	Exact. But routing shows variants to different visitors, so counted rates aren’t a fair comparison between variants.
Predicted	What the model believes now	A variant’s predicted rate, its difference from the default, signal, labels	Given with a 95% range, and moves as evidence arrives. It holds only as far as the model fits your visitors.
Measured	What a comparison set up by chance shows, without the model	The measured gain; a fixed share’s comparison of its variants	Given with a 95% range. The measured gain is the only figure that counts as a gain.

The command prints the kind beside each figure, or once for a block of them.

The control and the measured gain

The control is 10% of visitors by default, chosen at random, who always see the defaults, on every page and every visit. Everyone else gets routed visits. Since chance alone decides who is in the control, the two groups differ only by routing, and comparing them measures what routing changed.

A page’s measured gain is the extra conversions its routed visits produced, compared with showing everyone the defaults, with a 95% range: +246 conversions (signup) vs defaults [+9, +482].

It counts matured visits only, never projecting visits whose window is still open. The maturity beside it is the share of its visits that have matured.
It measures routing as it runs, exploration and the cautious start of new variants included.
A promote keeps it. The gain so far is kept, and measurement continues against the new defaults, which the control now sees. So it adds up what routing added over each period’s defaults.
Each page’s gain stands alone. Visitors see several pages, so gains don’t add up to a site total.
It can be negative, and stays so in the total.
The range is not a stopping rule. It is recomputed every day, and acting on the first good day overstates gains.

The control should get its share of new visitors. When a page’s control drifts from it, often because of caching or the proxy’s matcher, the control alarm warns that the measured gain may be biased.

Ranges

Every estimate comes with a 95% range: the values the evidence supports. Say the range with the value, as “+16%, somewhere between +1% and +33%”.

A range that spans zero means the direction isn’t known yet. −33 conversions (signup) vs defaults [−218, +152] could be a loss or a gain.
A range entirely above or below zero shows the direction; the size can be anywhere in the range.
No range shows while there are too few visits; the value and its counts still show.

Signal

Signal says how much evidence backs a variant’s predicted rate, from 0 to 100, labeled New, Early, Building or Strong. It grows with the variant’s own matured visits; evidence borrowed from a parent alone never makes it strong. It is not a probability of being best.

Labels

A label says what routing is doing with a variant now:

favored: routing intends to show it most in its slot, ties included;
learning: routing has learned from its own visits;
prior: not yet, so its prediction comes from the page’s other variants, or from its parent.

capped or floor-supported beside a label means a bound is holding its share down or up: one you set, or one routing applies itself, as to a new variant. default marks the slot’s default; it isn’t a label. No label is a verdict: what to keep is a judgment on the evidence, yours or your agent’s.

Scorecards

A scorecard opens when a change goes live, and compares:

a new variant with its slot’s default, at the same time;
an edited variant’s new version with the version it replaced, over their own periods, so not at the same time;
a resumed or reverted variant with its slot’s default, afresh;
every live variant with its default, afresh, when the page’s objective changes.

Its result is predicted: routing, not chance, chose who saw each side, so the comparison rests on the model, which the output marks as served. A slot with a fixed share also gets a measured comparison of its variants, over the visits split between them. A scorecard closes when one side stops being served, after a pause, a retirement or a newer edit, and keeps its last values. A promote closes none.

What updates when

What	When
A deploy or an act reaching visitors	Within about 2 minutes
Live activity	Within a minute
Counts, and most warnings	Every hour; every 15 minutes on Pro
Learning	Once a day; on Pro, as conversions mature, at most hourly
The measured gain	Once a day
Predictions	After each fit of the model, and soon after each deploy or act

Each output says how fresh it is. In counts to 09:00Z · estimates matured to 10-11, counts include what arrived by 09:00 UTC, and estimates use the visits matured by October 11. Days are UTC days. A prediction waiting to be recomputed after a deploy or an act keeps its last value, marked (stale).

When a value is missing

A missing value says why, and none of these means zero:

The output says	It means
`not prepared yet`	The first counts refresh or daily preparation it needs hasn’t run yet; the output says which.
`nothing yet`	There is no data for it yet.
`no prediction`	The page has had no traffic to predict from yet.
`no_model`	No prediction yet, since the model doesn’t cover it, as before the page’s first fit.
`not matured` or `not_matured`, with a day	No visit behind it has matured; the first ones do on that day.
`no range: too few visits`	The value and its counts show, with no range yet.
`no range`, on a percentage	The default’s own rate is too uncertain for a percentage; the difference in points shows instead.
`no_control` or `no_control_episodes`	There are no control visits, or none matured, so no measured gain yet.
`no_support`	The two sides were never shown to the same traffic, so they can’t be compared.
`(closed)`	A closed scorecard, at its last values.

Automated traffic

Crawlers, link previews, monitoring, ad checks and assistants fetching a page to answer someone are left out of learning and results, and counted apart: status gives their share of the month’s visitors on routed pages. Agents that browse for a person and can convert stay in, as does anything not recognized. A person who follows an assistant’s link is an ordinary visitor, whose arrival comes from that assistant.