I compared Tika with poi-3.15-beta1 vs the pre-release poi-3.15-beta3.
A number of exceptions were fixed. There was only one new exception.
There may be two small regressions in content:
1) some footers in PPT are not being extracted ("Prague" doesn't appear in -beta3)
2) some numbers in XLS are being corrupted
NOTE: these may be the fault of something we're doing at the Tika level. However, the upgrade from beta1 to the pre-release beta3 required no code changes.
More investigation is required.
The full batch of reports is available on github.
To download the original files, prepend: http://126.96.36.199/docs/