I am french speaking and I do not know you and there is nothing in your name that tells me you are a women. It might be the same for a lot of people too.
I stopped looking at your issue not because you are a women, as I wrote I did not know, but because you published synthetic data that did not matched the code you did not wanted to published at first. It is really hard to help somebody with the wrong data without seeing the code you claimed was perfect.
In my book the problem lies on the reading side of your application. You have confirmed that the data is written correctly because you see it in Compass. Then the problem lies in the reading side. I know the code is there now, but I have lost interest because of the hurdles, fake data unpublished code, you put on us. I do this on my free unpaid time and when there is too much obstacles, I work on something else. Not because of the person that ask the question or pose the problem.
Since I think it is best to published real documents when asking people to help, I recommend that you look at https://docs.mongodb.com/manual/reference/operator/aggregation/redact/. This way you may published real documents with sensible data redacted out. This way we can match the code with the data.
As I said scanning code for potential defects is time consuming. It is very frustrating when you realize that the code do not match the data because it is fake data.
You were redirected from the SERVER project in Jira because there wasn’t enough information for the team to investigate or suspect a possible server issue. There is no server-side caching of queries and your described scenario involves multiple drivers and applications interacting with data. If a minimal, reproducible test case can be created that clearly implicates server (or driver) behaviour, it would then be appropriate to create an issue in the relevant queue.
Assistance via a public forum requires a collaborative approach. We do not have access to your environment, your code, or your data and are trying to construct theories or reproduction cases based on the information provided. Someone asking for information may not be able to provide a solution, but additional details can be useful for others to provide suggestions and help you work through the problem.
The behaviour you are describing seems to be specific to your implementation (which you are understandably unable to share) and does not appear to be reproducible given the details you have provided so far. If the data is queryable as expected in Compass, my best guess would be an issue with your Node.js application. Since there was an issue with the synthetic example provided, I suggest asking a work colleague to review your application code in case there is something you may have overlooked.
If you need dedicated support or consulting time, there are commercial options which are outside the scope of this community channel.
Community members are volunteering their time to try to help you succeed. We expect all members to adhere to our MongoDB Code of Conduct including being positive, respectful, and considerate. Please take a moment to review these guidelines and assume best intent in your future interactions.