Test that long text is chunked and mean pooled.

Test that long English text is chunked by token budget.

Loopback addresses should be allowed for self-hosted scenarios.

Test Markdown detection with code blocks.

Test Markdown detection with headers.

Test Markdown detection with links and headers for stronger signal.

Test Markdown file extensions.

Passing migrate_to reassigns models instead of deleting them.

Test mixed config: generic + mode-specific (generic should enable all).

Test that mixed-language text is chunked using token measurement.

Test mode-specific env vars (LLM + EMBEDDING) enable only those 2 modes.

Test if a specific model is correctly configured and functional.

Model discovery should not append /models twice.

Test that each ModelManager instance is independent (not a singleton).

Test that openai-compatible is not available when no env vars are set.

Test files without extension.

Test that heuristics are used when no extension is available.

Test that missing model raises ValueError.

Test None input.

None title IS replaced by extracted title.

Test that result is normalized to unit length.

Test notes can hold content for embedding. Note: Embedding is now handled via command submission in Note.save(), not via needs_embedd

Test empty content is rejected.

Test archived flag defaults to False.

Test notebook context includes linked note content.

Test notebook context includes full source content for podcasts.

Test source context failures are not swallowed by notebook context.

Test notebooks with no source or note content produce empty context.

Test empty/whitespace names are rejected.

OpenAI-compatible provider URLs should be validated.

OpenAI model discovery should call the configured API base URL.

Test parsing single thinking block.

Test parsing with invalid input types.

Test that very large content is not processed.

Test parsing malformed output where opening <think> tag is missing.

Test parsing multiple thinking blocks.

Test parsing content without thinking tags.

The returned path must never contain user-supplied episode names. Since build_episode_output_dir does not accept an episode name at all,

Test PatternChainState structure and fields.

Placeholder 'Processing...' title IS replaced by extracted title.

Test plain text detection.

Test plain text file extensions.

Private IP addresses should be allowed for Ollama provider.

Private IP addresses should be allowed for self-hosted scenarios.

Test that prompt graph compiles correctly.

Public hostnames should be allowed.

Test that same instance is returned for same record_id.

Test removal of non-ASCII characters.

Test that pure ASCII text is unchanged.

Test removal of non-printable characters.

Test that newlines and tabs are preserved.

Test transformation raises assertion with no content.

Test that large Markdown sections are further chunked by tokens.

Test short text defaults to plain.

Test that short text is embedded directly without chunking.

Test that short text is not chunked.

Test mean pooling with single embedding returns normalized version.

Test RecordID parsing for command field.

Test that deleting a source removes the associated file.

Test that source deletion continues even if file deletion fails.

Test that deleting a source without a file doesn't fail.

Test speaker profile validates count and required fields.

Test notebook podcast jobs submit real source content, not model repr.

Test successful embedding generation with mocked model.

Test text within the token chunk size limit.

Test fallback when tiktoken raises an error.

Test fallback when tiktoken raises a network error (issue #264). In offline environments tiktoken.get_encoding() tries to download the

Test transformation model creation.

Test that transformation graph compiles correctly.

Test TransformationState structure and fields.

Test mean pooling with two embeddings.

Test unknown extensions return None.

Test that command_id is None on update when no embedding is triggered.

Test that updating a note returns the embed command_id.

Valid HTTP URLs should pass.

Valid HTTPS URLs should pass.

Test that vectorize() raises ValueError for empty string.

Test that vectorize() raises ValueError (not DatabaseOperationError) for empty text.

Test that vectorize() raises ValueError for whitespace-only text.