> ## Documentation Index
> Fetch the complete documentation index at: https://docs.surnex.io/llms.txt
> Use this file to discover all available pages before exploring further.

# How data collection works

> Which data is collected on a schedule, which runs on demand, and which is fetched live.

Almost nothing in Surnex is fetched when you open a page. Data is collected in the background, written to your project, and read back when you visit. This is why a page can show yesterday's figures and why a fresh project looks empty at first.

Understanding which mode a feature uses tells you what to expect when a number looks stale.

## The three modes

### Scheduled

Four job types have a schedule, and every project gets all four:

| Job                   | Feature                             |
| --------------------- | ----------------------------------- |
| `rank_check`          | [Rank tracking](/tracking/overview) |
| `backlink_snapshot`   | [Backlinks](/backlinks/overview)    |
| `geo_snapshot`        | [GEO visibility](/geo/overview)     |
| `local_ranking_check` | [Local SEO](/local/overview)        |

A new project gets all four **daily at 00:00 UTC**, enabled. Report delivery is scheduled separately, per report — see [Automate delivery](/reports/automation).

Each schedule has three parts:

| Part         | Meaning                                                             |
| ------------ | ------------------------------------------------------------------- |
| **Cadence**  | How often — daily, every 3 days, weekly, biweekly, or monthly       |
| **Run hour** | Which hour of the day, in the schedule's own timezone. Can be unset |
| **Timezone** | Any IANA timezone, and it **is** applied                            |

Cadence is a real interval — three days means three days, everywhere in the year. It isn't a day-of-week or day-of-month rule, so "weekly" is seven days from the last run rather than a fixed weekday.

Leaving the **run hour** unset makes the cadence pure elapsed time: the job runs every N hours from whenever it last ran, with no calendar involved.

The scheduler wakes every minute and claims whatever is due. Claiming a schedule advances its next run in the same statement, so a job can't be picked up twice — there's no dedup window to reason about and no "did this already run recently" guard.

### Collected on demand

These have no schedule. They run when you ask:

* [Site audits](/audits/run) — start a crawl from the Audits page
* [Web vitals](/web-vitals/overview)
* [AI search](/ai/overview) — AI Mode, ChatGPT visibility, benchmarking, citation gap
* [Domain overview](/domains/overview) and [tech stack](/domains/tech-stack)
* [Keyword research](/keywords/research)

<Note>
  Creating a project **does not collect anything immediately**. Its first jobs run when their schedules come due — which, on the default midnight-UTC schedule, may be up to 24 hours later. A new project looking empty is expected.

  Audits and web vitals never run on their own at all. A months-old audit score means nobody has crawled since.
</Note>

### Live

Two pages call out while you wait, with no stored history:

* [Explore trends](/trends/explore)
* [Trending now](/trends/trending)

These are the only pages where refreshing gives you newer numbers.

## Why a page is empty

| Cause                     | What you'll see                                     | What to do                                    |
| ------------------------- | --------------------------------------------------- | --------------------------------------------- |
| No scheduled run yet      | Empty state on a new project                        | Wait for the schedule, or lower the run hour  |
| It's an on-demand feature | Empty audits, web vitals, domain data               | Run it                                        |
| Nothing to collect        | No backlinks on a new domain, no local pack results | Nothing — the job succeeded and found nothing |
| The job failed            | Data for other features but not this one            | Check [Notifications](/notifications/view)    |

## One job at a time

The most expensive jobs refuse to start if one is already running for that project. Requesting a second [site audit](/audits/run) while the first is in flight is rejected rather than queued — a crawl is the slowest and costliest thing the platform does, and the slowest to notice you've started twice.

## Usage

Scheduled collection consumes plan quota the same way manual actions do. A daily rank check on 500 keywords draws on your keyword allowance every day, and lowering a cadence is usually the cheapest way to stay inside a limit.

[Usage and limits](/billing/usage) shows what's been consumed; [Plans](/billing/plans) lists the allowance on each tier.

This is why tracked keyword counts matter more than they look: each one is a recurring daily cost, not a one-off.