Recently, string extraction (done with Google's translation infra) failed a couple of times due to ICU msg format issues.  

In one case, the message in question was valid, but Google's translation infra does not support that particular format. 

In the other case, the message was an invalid ICU plural format. (bug 817450 ). 

If we add a presubmit check to validate ICU message format, we can discover this issue much earlier. 

I'll find out if it's possible to "extract" Google's internal ICU msg format checker and open source it.