@danielnaber @Yakov @marcoagpinto @matheuspoletto

Hi everybody,

Hope you are all having a great weekend.

Yesterday’s regression tests were supposed to be calm but they have shown odd results that I am still trying to diagnose.

https://languagetool.org/regression-tests/20161029/result_pt_20161029.html

Making my usual ‘find’ query to the rules with changes, I have gotten the following results:

```
ESTAR_SEGURO_QUE +5 (LEGIT)
VERBO_DE_QUE +1 (LEGIT)
ESTAR_CLARO_DE_QUE +3 (LEGIT)
INSISTIR_DE_QUE 0
PROFANITY 0 (1 HIDDEN)
3 words fixed -1 (6 matches in the same change description)
Marco rules 0
Ellipsis change pushed by Yakov:
UPPERCASE_SENTENCE_START -111 (MOST LEGIT NEGATIVES OR HIDDEN)
```

Great results… but:

The regression output has 15.148 lines.

Assuming that each change produces 15 lines (it is actually quite less) that would produce a regression test with roughly 1.815 lines (121 * 15).

The regression verbosity is at least 8 times larger than expected.

The issue seams due to former changes appearing again, in duplicate, as both new detections and as dropped detections.

I do not know how to understand these extra results.

Can anyone with more experience explain this extra verbosity in results and what can be done avoid it?

Cheers!

PS - Considering the hidden query results, the summary fits:

```
-Portuguese: 4661 total matches
+Portuguese: 4643 total matches
Portuguese: ø0,12 rule matches per sentence
4.661 + (5 + 1 + 3 - 1 - 111) = 4.559
```