Machine learning Alerts are failing for java error

Ok, the execute watch API output would be great, but we can also get some watch history output,

can you run (and likely put it into a gist, as it will contain 10 executions which hopefully contain stack traces)

GET .watcher-history-*/_search
{
  "query": {
    "bool": {
      "filter": [
        {
          "term": {
            "watch_id": "ml-cpu"
          }
        }
      ]
    }
  },
  "sort": [
    {
      "trigger_event.triggered_time": {
        "order": "desc"
      }
    }
  ]
}