I have noticed the exact same problem

No, I have tried. The time interval between API detection and manual detection is no more than 30 seconds, but the scores are very different. Later, I finally found that it should be the problem of the code. I used the “Java client library: bottometer4j by Samuel Toepke”. It may not get the relevant information of the user correctly, resulting in the difference in detection scores. After I replaced the “botometer-python library” it was normal
But what bothers me is that due to the daily limit of the API, it means that the data that I spent more than 20 days detecting needs to be re detected. Moreover, I notice that the “Java client library: bottometer4j by Samuel Toepke” was last updated a few years ago. I hope the official can fix it in time or explain that it has been abandoned, otherwise it will mislead more people

Thank you for using our tool! Both the webpage and the API use the same machine learning model. However, the scores reflect the status of the accounts at the moment when they are checked. So it’s possible for Botometer to return different scores for the same account if you check then at different time.

–Botometer Team