grobidOrg / grobid / 29392540579

39%

master: 43%

Pull #1506

github

Report per-dataset outcomes in run_evaluation.sh summary

The script folded every problem into a single boolean, so the final
'One or more evaluations failed' gave no clue which dataset failed or
why, even when the visible tail of the run was fully successful.

- track per-dataset outcomes and print an end-of-run summary naming
  each dataset with OK or FAILED (reason + log file to inspect)
- skip directories without PDFs during dataset discovery, so stray
  folders in the eval root no longer run a pointless evaluation and
  trip the failure flag through the 'report not found' branch
- remove stale report.md before each run so a leftover report can
  never be moved as if it were the current dataset's result
- fix REPO_ROOT to point at the repo root instead of grobid-home,
  making the default report source path actually exist

8920 of 25290 branches covered (35.27%)

Branch coverage included in aggregate %.

19072 of 45920 relevant lines covered (41.53%)

1.68 hits per line