SignalProcessor must not invoke the LLM it was constructed with — the rating is parseable from the rendered PM markdown directly.

Only the n_cross most recent cross-ticker entries are included.

More than 3 cross-ticker completed entries → only 3 injected.

Only the n_same most recent same-ticker entries are included.

More than 5 same-ticker completed entries → only 5 injected.

Coding Plan / MiniMax-Text-01 / any non-M2-prefixed model must NOT receive reasoning_split at all (top-level or extra_body) (#826).

The general-purpose NormalizedChatOpenAI must not carry DeepSeek-specific behaviour. Only the subclass does.

Labels should no longer claim '(local)' since the endpoint is dynamic.

Ollama users with custom-pulled models can pick 'Custom model ID'.

Skipping effort must not break other passthrough kwargs.

If a provider does not support with_structured_output, the agent falls back to a plain invoke and returns whatever prose the model pro

PM prompt omits the lessons section entirely when past_context is empty.

The structured PortfolioDecision is rendered to markdown that downstream consumers (memory log, signal processor, CLI display) can par

create_portfolio_manager accepts only llm; passing memory= raises TypeError.