Strange (at least to me) Variations in Consequtive Readings

[whitedavidp]

I am still in my initial phases of using this amazing package on my water meter. With the help of folks here, I am making progress. Thanks!

But I am getting some confusing readings. And they can occur and re-occur over consecutive cycles. Here is an except from my own logging of the mqtt json data. First off is that some of the messages sent via mqtt seem time lack timestamp data:

{"value": "","raw": "111315.65","pre": "111308.23","error": "Rate too high - Read: 111315.65 - Pre: 111308.23 - Rate: 7.42","rate": "","timestamp": "2025-05-27T18:02:14-0700"}
{"value": "","raw": "1N1315.59","pre": "111308.23","error": "Rate too high - Read: 111315.59 - Pre: 111308.23 - Rate: 7.36","rate": "","timestamp": "2025-05-27T18:02:14-0700"}
{"value": "","raw": "111316.69","pre": "111308.230","error": "Rate too high - Read: 111316.69 - Pre: 111308.23 - Rate: 8.46","rate": "","timestamp": ""}
{"value": "","raw": "111318.59","pre": "111308.230","error": "Rate too high - Read: 111318.59 - Pre: 111308.23 - Rate: 10.36","rate": "","timestamp": ""}
{"value": "","raw": "111318.60","pre": "111308.230","error": "Rate too high - Read: 111318.60 - Pre: 111308.23 - Rate: 10.37","rate": "","timestamp": ""}

Next is that reading results can vary over cycles - even while our household is asleep and there is no water usage (and so consecutive images should be the same - which I have not yet confirmed, only inferred), and the conditions (light, etc) out in the crypt by the street where the meter lives do not change:

{"value": "","raw": "11131N.38","pre": "111308.38","error": "Rate too high - Read: 111318.38 - Pre: 111308.38 - Rate: 10.00","rate": "","timestamp": "2025-05-28T03:14:11-0700"}
{"value": "","raw": "111316.38","pre": "111308.38","error": "Rate too high - Read: 111316.38 - Pre: 111308.38 - Rate: 8.00","rate": "","timestamp": "2025-05-28T03:14:11-0700"}
{"value": "","raw": "111316.38","pre": "111308.38","error": "Rate too high - Read: 111316.38 - Pre: 111308.38 - Rate: 8.00","rate": "","timestamp": "2025-05-28T03:14:11-0700"}
{"value": "","raw": "111316.38","pre": "111308.38","error": "Rate too high - Read: 111316.38 - Pre: 111308.38 - Rate: 8.00","rate": "","timestamp": "2025-05-28T03:14:11-0700"}
{"value": "","raw": "111316.38","pre": "111308.38","error": "Rate too high - Read: 111316.38 - Pre: 111308.38 - Rate: 8.00","rate": "","timestamp": "2025-05-28T03:14:11-0700"}
{"value": "","raw": "11131N.38","pre": "111308.38","error": "Rate too high - Read: 111318.38 - Pre: 111308.38 - Rate: 10.00","rate": "","timestamp": "2025-05-28T03:14:11-0700"}
{"value": "","raw": "111316.38","pre": "111308.38","error": "Rate too high - Read: 111316.38 - Pre: 111308.38 - Rate: 8.00","rate": "","timestamp": "2025-05-28T03:14:11-0700"}
{"value": "","raw": "111306.39","pre": "111308.38","error": "Rate too high - Read: 111306.39 - Pre: 111308.38 - Rate: -1.99","rate": "","timestamp": "2025-05-28T03:14:11-0700"}
{"value": "","raw": "111316.39","pre": "111308.38","error": "Rate too high - Read: 111316.39 - Pre: 111308.38 - Rate: 8.01","rate": "","timestamp": "2025-05-28T03:14:11-0700"}
{"value": "","raw": "111316.39","pre": "111308.38","error": "Rate too high - Read: 111316.39 - Pre: 111308.38 - Rate: 8.01","rate": "","timestamp": "2025-05-28T03:14:11-0700"}
{"value": "","raw": "111316.39","pre": "111308.38","error": "Rate too high - Read: 111316.39 - Pre: 111308.38 - Rate: 8.01","rate": "","timestamp": "2025-05-28T03:14:11-0700"}
{"value": "","raw": "111306.39","pre": "111308.38","error": "Rate too high - Read: 111306.39 - Pre: 111308.38 - Rate: -1.99","rate": "","timestamp": "2025-05-28T03:14:11-0700"}
{"value": "","raw": "11131N.39","pre": "111308.38","error": "Rate too high - Read: 111318.39 - Pre: 111308.38 - Rate: 10.01","rate": "","timestamp": "2025-05-28T03:14:11-0700"}

While my career was in programming and software engineering, I know almost nothing at all about AI systems. However, I have long come to expect that computers produce consistent and re-producible results - given constant inputs. I have seen but not yet logged, other much larger variances. So this is curious...

I am using the SybexXx models:

And I am currently using greyscale and some other settings that were recommended to me in another thread recently. Of course, those too did not change at 3am :-)

Suggestions? Thanks!