Discrepancy in execution units returned by eval

[ricomiles]

We’ve been working on finalizing eval_tx for Dolos, and we’re using the eval function to evaluate transactions. However, we’ve encountered an issue with the execution units being returned.

We’re running our tests in preview mode and using Blaze to test the transactions. Here are the results:

using u5c provider(with our eval_tx implementation): [19724_1, 5913986_2]
using other provider(kupmios): [19728_1, 6218182_2]

As you can see, there’s a small difference in the execution units. At first glance, it doesn’t seem significant. However, when we attempt to submit the first result using a different submitter (in this case, via cardano-cli), we receive the following error:

The machine terminated part way through evaluation due to overspending the budget. 
The budget when the machine terminated was:({cpu: -304196\n| mem: -4})
Negative numbers indicate the overspent budget; 

We’ve discovered that increasing the execution units to match or exceed those returned by kupmios allows the transaction to be submitted successfully. Our findings suggest the built-in cost models used by uplc might be outdated or not fully aligned with current ledger requirements.

[rvcas]

This crate is not yet recommended for production use