results-analysis
results-analysis copied to clipboard
Fix #88: Mark as bad any run with >=66 missing results
This is a high enough threshold that it doesn't exclude runs with missing results (e.g., due to -print or -crash being unsupported), but does catch the vast majority with an unusually high number of missing results.
https://gist.github.com/gsnedders/676cd70e21bcd8d96a79550582d455b3 for a diff of the output.
Oh, actually, maybe my data is bad here; seems to omit cases where entire trees are missing.
@gsnedders what are some of the runs that you're hoping this will exclude? Any idea what went wrong with them?
I would quite like it if we used the manifest when processing results in wpt.fyi, so that we could tell if a run is complete or not and label incomplete runs, or at least know for sure how many tests of which type are missing.