Live Test vs .net solution

I understand the system is still under development, and I expected there might be some bugs or limitations with new tools. However, I was under the impression that at least version 2 would perform at the same level as version 1 — if not better. Unfortunately, that hasn’t been the case so far.

In our experience, version 2 has shown regressions in areas that previously worked well, which is affecting reliability and confidence in the upgrade.

Tipically we are facing issues with:
1) Type casting - string for integer;
2) Value recognition: sometimes the system correctly interprets the comma as a decimal separator (e.g., "12,50" as twelve point five), but other times it does not, treating it incorrectly or ignoring it altogether. This inconsistency is affecting the accuracy of the extracted numerical values. In Portugal sometimes we use comma as separator and point, and this happens in both cases. (example: sometimes lets say Quantity Field 2,000 is taking 2.0 but sometimes in other documents it is treating 2000.0)
3) We’ve observed inconsistent results when scanning the same document multiple times. In some cases, the first scan successfully extracts all values, while a subsequent scan fails to do so, resulting in missing or incomplete data. In other cases, the opposite happens — the first scan misses data, but the second one captures it correctly.

In short, the output is not consistent between identical scans of the same invoice, even though the input remains unchanged. Comparing to Live Test also the results are inconsistent.
4) Most of the documents, Product code is coming null eventhough we have a value in document in the solution but it’s being recognized in green in Live Test.

5) For some of the documents, customer company registration is coming empty.
6) I also opened a different ticket here, reporting another strange case, where it recognizes the product code in first and third page but not in second.

Thank you, any help is appreciated.