Reliability report for the second quarter

<p>We publish a reliability report every quarter because a status page tells you what is broken right now and nothing about what a normal month looks like. Here are the numbers for April, May and June 2026.</p><p>The flow runner was available 99.94 percent of the time, which is 26 minutes of degradation across the quarter against a target of 99.9. The API sat at 99.98 percent. The median gap between a source event and the matching write at the destination was 41 seconds, and the ninetieth percentile was 3 minutes 12 seconds, which is what you would expect given that most source apps are polled every three minutes. Across the quarter we moved 214 million records and 0.31 percent of runs ended in an error, roughly two thirds of which were rejected by the destination app rather than by us.</p><p>One incident is worth describing. On 14 May a schema change on our side made the queue writer reject any record with an empty attachment array, and for 71 minutes about 9 percent of runs failed with an unhelpful validation message. Nothing was lost, since failed runs stay replayable, but 4,180 records sat waiting and a lot of people saw a red badge on a flow that was fine. We fixed the writer, replayed the backlog automatically, and have since added a canary flow that exercises every field type against a test workspace before a deploy reaches production.</p><p>The unhelpful message is the part that annoyed us most. Three of the eleven support tickets during that hour were people rebuilding a working field map because the error read as their mistake. Error text on runner failures now names the component and states plainly whether the cause was your configuration or ours.</p><p>Third quarter targets stay at 99.9 percent for the runner and a median under 60 seconds. The status page carries the live numbers and a ninety day history, and every incident write-up lands there before it appears here.</p>