The single longest incident StatusBird recorded in the last 90 days ran 15636 minutes: Cloudflare went non-operational on October 5, 2026 at 16:34 UTC and did not return a clean reading until October 16, 2026 at 13:10 UTC. The next four longest were Plaid at 1322 minutes, two separate Shippo incidents at 560 and 512 minutes, and OpenAI at 382 minutes, out of 45 major or critical incidents in the window.
StatusBird checks each of the services it tracks every 2 minutes, and every figure below comes from those readings over the 90 days ending October 16, 2026. Details of how checks and severities are recorded are on the methodology page.
Which incidents ran longest?
| Service | Started (UTC) | Recovered (UTC) | Duration | Worst status |
|---|---|---|---|---|
| Cloudflare | October 5, 2026 at 16:34 | October 16, 2026 at 13:10 | 15636 minutes | major |
| Plaid | August 2, 2026 at 14:52 | August 3, 2026 at 12:54 | 1322 minutes | major |
| Shippo | July 30, 2026 at 04:42 | July 30, 2026 at 14:02 | 560 minutes | major |
| Shippo | October 1, 2026 at 04:42 | October 1, 2026 at 13:14 | 512 minutes | major |
| OpenAI | October 7, 2026 at 05:32 | October 7, 2026 at 11:54 | 382 minutes | major |
For scale, the sixth longest was ShipStation at 254 minutes on September 25, 2026, starting at 07:57 UTC and recovering at 12:11 UTC. Everything below that mark in the 90-day window finished inside four hours.
Cloudflare: 15636 minutes, and it was still the open incident on October 16
This one is different in kind, not just in degree. The incident opened at 16:34 UTC on October 5 and the recovery reading landed at 13:10 UTC on October 16, which is the last day of the measurement window. Across 90 days Cloudflare accumulated 15754 min of downtime in total against a 90-day uptime of 87.82%, a grade of F, and 5 major/critical incidents. The second Cloudflare incident in the recent set was much shorter: 12 minutes on October 2, 2026, from 21:48 UTC to 22:00 UTC.
What a store would have seen depends entirely on which Cloudflare surface it sits behind. The worst status recorded was major, not critical, so this was a degraded reading rather than a flat refusal across the whole window. In practice stores with Cloudflare in front of their domain saw intermittent edge errors, slow first-byte times on cold pages, and image or script requests failing while the HTML loaded. That failure mode is the one most often misdiagnosed as a theme or app problem, because the storefront still renders.
Cloudflare also carries the highest current risk level StatusBird is showing, at a modelled 90.0% chance of a non-operational reading in the next 24 hours. The live view is on the Cloudflare status page, with the longer record on its reliability history.
Plaid: 1322 minutes across an overnight window
Plaid went non-operational on August 2, 2026 at 14:52 UTC and recovered on August 3, 2026 at 12:54 UTC, running 1322 minutes straight through the night. Plaid's 90-day uptime is 98.87%, grade C, with 1460 min of downtime from 3 major/critical incidents and an average incident length of 486 min.
A store using Plaid for bank account verification on ACH or pay-by-bank checkout would have seen verification attempts hang during the whole window. The symptom is not an error page: it is customers reaching the bank-link step and stalling, then abandoning. Because the incident opened mid-afternoon UTC and closed the following midday, a store owner who only checked dashboards in their own business hours could have logged in to a working Plaid and seen nothing wrong, while a full overnight run of attempts had already failed. The two more recent Plaid incidents were far shorter: 94 minutes on September 26, 2026 from 03:41 UTC, and 44 minutes on October 11, 2026 from 08:25 UTC.
Shippo: two long incidents that both began at 04:42 UTC
Shippo supplied two of the five longest incidents in the window, and the start times are identical to the minute. The first opened July 30, 2026 at 04:42 UTC and recovered at 14:02 UTC the same day, 560 minutes. The second opened October 1, 2026 at 04:42 UTC and recovered at 13:14 UTC, 512 minutes. Shippo's 90-day totals are 98.6% uptime, grade C, 13 major/critical incidents, and 1808 min of downtime, the second highest downtime figure of any service StatusBird tracked in the window.
Both windows cover the hours when most North American warehouses are not yet staffed and most European ones are starting. A store would have seen label purchases failing and rate requests timing out at exactly the point where the morning batch normally goes out, then a pile-up of unfulfilled orders waiting when staff arrived. The October 1 incident ended at 13:14 UTC, which for a US East Coast operation is mid-morning: the queue was still there, just purchasable again. Shippo's other recent incidents were short by comparison, including 130 minutes on October 8, 2026 from 21:44 UTC, 30 minutes on September 24, 2026 from 19:25 UTC, and 12 minutes on September 17, 2026 from 04:49 UTC. The pattern is on the Shippo reliability history.
OpenAI: 382 minutes on a single morning
OpenAI opened a major incident on October 7, 2026 at 05:32 UTC and recovered at 11:54 UTC, 382 minutes. It was not isolated: a second long one ran 206 minutes from October 9, 2026 at 21:01 UTC to October 10, 2026 at 00:27 UTC, plus 52 minutes on October 1, 2026 from 18:48 UTC and 56 minutes on September 25, 2026 from 22:59 UTC, the only incident in the recent set where the worst status recorded was critical. Across 90 days OpenAI sits at 97.89% uptime, grade D, with 696 min of downtime from 4 major/critical incidents.
Stores that route support replies, product description generation, or on-site chat answers through OpenAI would have seen those features time out or fall back to an error for most of that October 7 morning. The visible symptom is usually a chat widget that accepts a message and never answers, which customers read as being ignored rather than as an outage.
What the five timelines have in common
- Four of the five opened outside normal working hours for at least one major market. The two Shippo incidents began at 04:42 UTC and OpenAI's longest at 05:32 UTC. ShipStation's 254 minute incident began at 07:57 UTC.
- Length does not track severity labels. Every one of the five longest carried a worst status of major, while the only critical reading in the recent set lasted 56 minutes.
- Two services accounted for most of the long minutes. Cloudflare's 15754 min and Shippo's 1808 min dominate the 90-day downtime totals, and both show up repeatedly in the outage index.
- Repeat offenders repeat at similar clock times. Shippo's two longest both started at 04:42 UTC, which is a useful thing to know when you schedule your label batch.
Concrete changes these timelines justify
- Move your morning label batch off the 04:42 to 13:14 UTC band if you can. Both long Shippo incidents sat inside it. If you cannot move the batch, keep a second label source configured in your shipping settings so you can switch providers without a new account signup mid-incident.
- Set alerts on the dependency, not the dashboard. The Plaid incident ran 1322 minutes and closed at 12:54 UTC. Anyone checking at 09:00 local time the next morning in the Americas would have seen a healthy service and a bad night of conversions.
- Separate edge failures from app failures. During a Cloudflare major window the HTML often loads while assets fail. Before you roll back a theme change, open your Cloudflare reading and check whether the edge was non-operational at that timestamp.
- Give AI-backed chat a static fallback message. A 382 minute window of silent timeouts reads to customers as a store that ignores them. Configure the widget to show a plain contact option when the model call fails.
If you want to be told when one of these windows opens instead of reconstructing it afterwards, StatusBird checks these services every 2 minutes and sends SMS, email, Slack, Teams, or Discord alerts. You can set it up at statusbird.io/register.
FAQ
What was the longest single outage StatusBird recorded in the last 90 days?
Cloudflare, at 15636 minutes, from October 5, 2026 at 16:34 UTC to October 16, 2026 at 13:10 UTC. Its 90-day uptime across the window is 87.82%.
How long did the Plaid outage in August last?
1322 minutes. It started August 2, 2026 at 14:52 UTC and recovered August 3, 2026 at 12:54 UTC, with a worst recorded status of major.
Why does Shippo appear twice in the longest incidents list?
Shippo had two long incidents that both began at 04:42 UTC: 560 minutes on July 30, 2026 and 512 minutes on October 1, 2026. Over 90 days it recorded 13 major/critical incidents and 1808 min of downtime at 98.6% uptime.
Were any of the five longest incidents marked critical?
No. All five carried a worst recorded status of major. The only critical reading in the recent set was a 56 minute OpenAI incident on September 25, 2026 starting at 22:59 UTC.
Which service is most likely to go non-operational next?
Cloudflare, at a modelled 90.0% chance of a non-operational reading in the next 24 hours, followed by BigCommerce at 68.6% and Vercel at 41.5%.