Problem

If task has 2 segments like:

Segment 1 has size of 3 cm
Segment 2 has size of 4 cm

And system has integer rule active for language pair of task.

And task has t5memory language resource connected while TM has 1 segment:

EN: Segment 2 has size of 4 cm
DE: Segment 2 ist 4cm groß

IMPORTANT: 4cm has no whitespace.

Then when such a task will be pre-translated in master - 1st segment will be pre-translated with <100 match and repetition (2nd segment) will be left not translated.

The cause for this is integers in task will be protected by integer rule, but in TM only first integer will be protected:

EN: Segment <t5:n n="2" /> has size of 4 cm
DE: Segment <t5:n n="2" /> ist 4cm groß

This will happen cause 4cm has no whitespace and 4 cannot be recognised as int here.
Then pair can't be made for source and target protected content and 4 in source will be saved unprotected.

Then in pre-translation second segment will be considered repetition and logic will try to apply target for 1st one as translation for it.
But this will result in error with tags (1 less) and as we have bug match from TM for 2nd segment (100 match, guessed) will be returned with match-rate = 0.

and as end result 2nd segment will be untranslated.

Solution

Check if match-rate for current repetition from TM is not worse then match-rate of master and if so - use repetition match.

If master looks like a better choice (this is possible with guessing of match in t5memory, simply change order of segments in task above) - try to apply master match and fallback to repetition own match if not possible.