[J0s3f]

I used the tool on a mid size German Discord server and found a few issues. I fixed them:

Text messages without recognized sentiment terms currently serialize an unavailable score through NaN as byte zero. Both message readers turn that into -128, so an export with many unscored messages can appear overwhelmingly negative. This change preserves score availability through serialization and shows unscored coverage explicitly.

• Reserve byte zero for unavailable sentiment, keep balanced zero distinct, and preserve the sign of fractional emoji scores. Both readers use the same decoder; the byte layout is unchanged.
• Match words case-insensitively, consume the longest phrase once, and improve nearby negation with common German forms, a small set of dialect aliases, contrast boundaries and the not only / nicht nur exception. Matching no longer allocates a sliced token array at every word.
• Include unscored messages in the overview and percentage denominators, show coverage, and use a non-interactive chart legend so small categories remain readable without changing the denominator.
• Add regression coverage for serialization, both readers, complete database/report aggregation, phrase matching and negation. Document limitations and how to regenerate older standalone reports.
• Improve the German word list for chat: ignore discourse particles that carried polarity ("ja", "bitte", "sicher"), treat price words as complaints, add common German/Austrian profanity and complaint words, and keep negated insults negative ("ned so deppat"). - This is a separate commit and can be removed, if only the bugfix is wanted.

Compatibility: legacy byte zero cannot distinguish an unavailable score from a genuine score saturated at -128. Regenerating from the export resolves that ambiguity. New valid scores clamp to [-127, 127].

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.