In the FAQ, document/answer the question "Should I use one repository or separate repositories when backing up multiple clients/servers?". Provide some rationale and use case examples.

If somebody has some time for FUSE benchmarking:

diff --git a/src/borg/fuse.py b/src/borg/fuse.py
index 429790e4..27ab1c1a 100644
--- a/src/borg/fuse.py
+++ b/src/borg/fuse.py
@@ -644,12 +644,13 @@ def read(self, fh, offset, size):
                 data = self.data_cache[id]
                 if offset + n == len(data):
                     # evict fully read chunk from cache
-     

This thing works, but reading the Yarn documentation seems to suggest it shouldn't be necessary. I've tracked a few issues in the Yarn repo that suggest potential root causes, but I'm sure the authors of this package would be in a better position to explain it 😄

When disabled, accept stdin as-is and don't re-open the TTY. When enabled, accept a user-specified file separator sort of like how cut -d works.

The classifier LogisticRegressionClassifier.prob returns non-match probability instead of match probability. Is this dataset dependent behavior?

Looking at the source, setting -1 to either the size or content flags completely ignores potential changes in the size or content.

The help text implies that setting to -1 is equivalent to setting to zero (i.e. no difference in size/content allowed), whereas it's actually the opposite (any difference in size/content allowed).

Labeled training example pairs should be stored in a table for selection and reuse. Data stored for examples should include:

• Source
• Source ids
• Label
• Label date
• Comment to store labeling rules applied by labeler

Storing examples like this allows them to by reused in the following ways:

• Select specific subsets of labeled pairs to build models from
• Store multiple labels f