Availability and changes.

Measured since 27 Aug 2026. Time before that does not count as downtime: it was not measured at all.

Event ingestion

95.64 %

71,017 of 74,253 expected samples succeeded

From outside 100.00 %

Processing and attribution

95.72 %

71,076 of 74,253 expected samples succeeded

From outside not measurable

Customer portal

95.69 %

71,054 of 74,253 expected samples succeeded

From outside 100.00 %

Read API

95.72 %

71,078 of 74,253 expected samples succeeded

From outside 100.00 %

What is measured is the environment at bench.metrunio.com.

The large figure is measured from the inside: one sample per component per minute, taken by our own processing. If that fails too, no “unavailable” sample is written, nor any other. That is why we calculate against the expected number of samples, so a gap counts as downtime. This measurement sees what nobody outside can: the backlog in the queue. That is why processing carries no second figure.

The smaller figure below it is measured from the outside: an independent service calls our addresses every 5 minutes from several parts of the world and counts a minute as available only once the majority of locations got through. It sees what you see: network, load balancer and certificate included. What it still does not answer: whether the path out of your own network was clear. Where the two figures differ noticeably, the cause lies between the services rather than inside them.

There was no outage in the measured period. That is a statement about the samples that exist, not about time that was never measured.

Listed are outages a sample recorded. If the measurement fails along with the service, no “unavailable” sample is written, nor any other. Such an outage lowers the rate above but does not appear here.

• Event ingestion is the critical part: if it fails, measurement time is lost. It runs separately from the portal and reporting: an error in a report does not stop the measurement.
• An outage in processing causes delay, not loss. Events wait in a queue and are caught up.
• Maintenance is usually announced seven days ahead and scheduled into low-traffic windows. The terms guarantee at least 48 hours’ notice (§ 9).

A figure that moves between two weeks has a reason either in your data or in our software. So that you can tell the two apart, this is what we changed, including what was wrong before.

1. 26 Sept 2026

   New

   The privacy notice now explains what happens when you sign in with Google Ads or Meta: which data we read from the ad account, what for and how to revoke access. It also states that data from Google APIs is subject to the Limited Use requirements.

2. 29 Aug 2026

   New

   Yearly prepayment: twelve months for the price of ten. The monthly plan stays cancellable monthly.

3. 29 Aug 2026

   New

   Undeliverable addresses are recognised and receive no further mail. If you expect a notification and none arrives, the reason is in the portal.

4. 29 Aug 2026

   New

   The trial now actually exists: 30 days of Growth from sign-up, with reminders seven days and one day before it ends. After that the account continues on the free plan.

5. 29 Aug 2026

   Fix

   Post-view and the conversion pushback now apply only in the plans that include them. Until now every account received them, including the free one, where the pricing page ruled them out.

6. 29 Aug 2026

   Fix

   The guest access on the demo page now points at the curated example account and shows credentials that actually work. Before, it led nowhere.

7. 25 Aug 2026

   New

   Two-step verification for every account, with recovery codes.

8. 25 Aug 2026

   New

   Audit log in the portal: who changed what and when, with the reason.

9. 25 Aug 2026

   New

   Raw data view. Every metric links to the events behind it.

10. 25 Aug 2026

    New

    The journey detail view shows the share each model assigns to every touchpoint.

11. 25 Aug 2026

    New

    Attribution windows and the default model are selectable in the settings.

12. 25 Aug 2026

    Fix

    Attribution is ready again markedly faster after a change. The figures do not change; they were correct before, too.

13. 25 Aug 2026

    Fix

    The model comparison showed the total attributed conversions, which is necessarily identical in every model. It now shows the distribution across channels, which does differ.

14. 25 Aug 2026

    Fix

    Lists in the portal ended after 25 rows with no pagination. Journeys, campaigns, conversions, imports, exports, keys and privacy requests can now be paged and sorted.

15. 25 Aug 2026

    Fix

    Every diagnostic finding linked to a guide that did not exist. Every error code now has its own page.

16. 25 Aug 2026

    Limit

    The export limit is 200,000 rows instead of 5,000. For the full dataset the read API with pagination is still the way.

17. 24 Aug 2026

    New

    Alerts for missing data, a drop in conversions and an exhausted allowance.

18. 24 Aug 2026

    New

    Help centre with search, machine-readable output and one source for website and portal.