The vocabulary, defined

Terms

AI crawler

A bot operated by an AI provider that fetches your pages so an assistant can index or cite them. It is machine consumption of your content, never a visitor. Counting it as traffic growth turns machine reading into a growth story with no people in it.

AI referral

A person arriving from an assistant that cited or recommended you, identified by the referrer on the arriving request. This is acquisition, not machine reading. Separating it from search shows whether assistants send you customers or only read your pages.

Anomaly

A measurement outside its expected range once seasonality, weekday pattern and normal variance are accounted for. An anomaly is not automatically a problem, and not every problem is an anomaly — a steady decline can sit inside the band for weeks.

Attribution

A rule for assigning credit for a conversion to the touchpoints before it. It is a model, not a measurement: last click, first click and position-based rules produce different answers from identical data. The rule is always shown beside the number.

Attribution window

The period before a conversion within which an earlier visit is still considered part of how it came about. It is a rule you choose, not a fact discovered in the data, and changing it changes every credited figure — which is why Reveliqo reports the window alongside the number.

Bot

Any non-human client: crawlers, monitors, scrapers, scanners, link previewers and attack traffic. Bot is a category, not an accusation — a monitoring check and a credential-stuffing attempt are both bots, and they are kept in separate classes.

Business event

An outcome your own server reports — a lead, a trial, an account, an order, a subscription change or a refund. It is the authoritative record of a conversion, because it comes from the system that actually performed it rather than from a browser that may have closed, blocked the script or retried the request.

Causation

One thing actually producing the change in another. Establishing it needs a controlled comparison — an experiment, a holdout, a staged rollout — not a matching timestamp. Reveliqo states the correlation it found and names what would be required to prove cause.

CLS

Cumulative Layout Shift: how much content moves unexpectedly while a page is being read. It is the metric behind a mistapped button when a late font or an advert pushes the layout down. Zero is stable, and the number has no unit.

Conversion

A business outcome you defined: a lead, a trial, an order, a subscription, an upgrade. It is a decision about what matters, not a technical detail. Server-side conversions are taken as the record, because a browser event can be blocked, retried or lost.

Core Web Vitals

Three field measurements of loading, responsiveness and visual stability: LCP, INP and CLS. Smart Beacon collects them from real browsers, so they are attributable to a page, a device class, a country and a release rather than to a single score.

Correlation

Two things moving together over the same period. It is evidence worth investigating, and it is the strongest statement a timeline alone supports. Reveliqo writes “the change begins shortly after” precisely because that is what was observed.

Crawler

An automated client fetching pages on a schedule, usually for a search index. Verified crawlers are confirmed by reverse lookup and published address ranges rather than taken at their word, because a user agent string is a claim, not an identity.

Edge and origin

The edge is where a request first arrives on the network and is recorded. The origin is your own server or hosting, which the request then reaches unchanged. Reveliqo reads what the edge already observed; it adds nothing to the page and puts no analytics call inside the path between the two.

Engagement time

How long a page was actually in the foreground and being interacted with, reported by Smart Beacon. It replaces time-on-page arithmetic, which counted a tab left open overnight as engagement and counted the last page of a visit as zero.

Event

A single recorded thing that happened, with a timestamp and properties: a request, a click, a form submit, a trial started, an order refunded. An event is not automatically a conversion — a conversion is an event you decided counts.

Idempotency key

A value you attach to a business event so the same event can be sent twice without being counted twice. Webhooks retry, deploys replay queues and networks drop responses; without the key, one recovered retry quietly adds a second sale to your revenue.

Identity mode

The collection setting that decides how much a visitor can be recognised: Minimal keeps no persistent identifier, Visit recognises a visitor within one visit, and Customer links visits to a known customer through a business event. It governs what every report can answer, and choosing it — within what your legal position and consent state allow — is the customer's decision, not the product's.

INP

Interaction to Next Paint: the delay between a person interacting and the page visibly responding, measured across the whole visit. It replaced First Input Delay because responsiveness after the first tap is what people actually notice and complain about.

LCP

Largest Contentful Paint: how long until the largest element in the viewport has rendered. It is a proxy for the page looking ready, and it is reported at the 75th percentile because an average hides the quarter of visits that were slow.

Pageview

One document request rendered for a person. It is a unit of consumption, not of interest or value. A page can lead every pageview ranking and produce almost no revenue, which is why content is also ranked by revenue per visit.

Person (visitor)

A distinct human counted within a chosen window, after crawlers and bots are removed. It is not an identified individual, and it is not one person across every device they own — a phone and a laptop count twice unless a business event ties them to one customer.

Real-user monitoring (RUM)

Performance measured from the browsers of actual visitors, on their devices and their connections, rather than from a test agent. It reports the distribution you really serve, including the slow tail that synthetic testing never sees.

Referrer

The site a request says it came from. It is reported by the browser or the request itself, so it can be absent, stripped by a privacy policy, or rewritten by a redirect. An empty referrer means unknown origin, not direct traffic.

Revenue per visit

Revenue divided by visits for the same segment and period. It ranks a channel by what it is worth rather than by how much it sends. It is not profit: costs are not in it unless a finance system supplies them.

Route template

The pattern behind a set of URLs — /product/[id] rather than ten thousand individual product pages. Performance, errors and conversions are reported per template so a regression on one route is visible instead of being spread thinly across every URL that uses it.

Semantic metric

A named measure with one stored definition — people, conversion_rate, revenue_per_visit — used by dashboards, exports, the API and every AI answer. The AI selects from these definitions; it does not create a measure of its own, which is why an answer, a dashboard tile and an export agree.

UTM parameters

Campaign labels added to a link: source, medium, campaign, term and content, read from the request itself. They describe what you tagged, not where someone actually came from — an untagged link and a mistyped label both end up filed elsewhere.

Visit (session)

One continuous stretch of activity by the same person, ending after a period of inactivity or at a defined boundary. A visit is not a visitor: somebody reading you on Monday and again on Thursday is one person and two visits.