Files Going to Inbox

Why are files ending up in 0_Inbox/ instead of correct categories?

Reasons for Inbox Classification

  1. No signals matched - All 6 signals failed
  2. Low confidence - Below threshold (default 0.75)
  3. Unrecognizable content - Not enough text to match
  4. Ambiguous document - Doesn’t fit any route well

Solutions

1. Add Utterances

Help semantic matching by adding keywords:

# What should this document match?
uv run para-files add-utterance correct-route "describe the document"

# Re-classify
uv run para-files classify problem-file.pdf

2. Register Issuer

If from a known company:

uv run para-files add-issuer "Company Name" -c category

# Now matches with 90% confidence
uv run para-files classify document.pdf

3. Use Interactive Learning 

uv run para-files learn wrong-file.pdf

# Correct the classification
# Add keywords for future files

4. Add Confidence to Routes

Check if utterances are too weak:

# View current utterances
uv run para-files routes --utterances

# Add more specific ones
uv run para-files add-utterance route "very specific phrase"

5. Lower Threshold (Last Resort) 

export PARA_FILES_MLX_SCORE_THRESHOLD=0.70

# Only if other methods don't work
# Better to improve utterances instead

Debugging

See why a file went to Inbox:

# Verbose output shows all signals tried
uv run para-files classify problem-file.pdf -v

# Check confidence for specific route
uv run para-files test-route expected-route --file problem-file.pdf

# See what utterances exist
uv run para-files routes --utterances

Common Causes

Generic filename (invoice.pdf) → Add content-based utterances

Unknown company → Register as issuer or add utterances

Too few characters → Ensure 2000+ chars readable (try -v flag)

Empty or binary file → Can’t extract content, needs manual categorization

Prevention

  • Add utterances for each route before moving many files
  • Register known issuers early
  • Use learn regularly to build training data