v1.8.0

Added

• Added call flow pipeline: FlowCandidate and EntryPoint models in deepdoc/planner/flow_candidates.py trace endpoint families and runtime tasks/schedulers through the call graph to collect entrypoints, call chains, side effects (Celery/signal/event dispatches), and external touchpoints.
• Added _ensure_flow_buckets and _expand_flow_bucket_ownership in the planner to automatically create "Core Workflows" documentation buckets for the top-scored flow candidates and enrich their file/symbol ownership from call chain evidence.
• Added flow_context to AssembledEvidence; the evidence assembler renders a structured call flow section (entrypoints, call chain, side effects, external touchpoints) for any bucket with a flow_id generation hint.
• Added _check_flow_grounding validator that marks pages invalid when a flow bucket omits required "Call Flow" or "Side Effects" sections despite flow context being present.
• Added {flow_context} placeholder to all prompt templates in bucket_types.py and page_types.py.

Changed

• Nav sections proposed by the LLM are now preserved as-is rather than being replaced with hardcoded generic labels. _canonical_section_for_bucket only overrides empty or placeholder sections; supporting-tier buckets (tests, CI/CD) still get properly re-sectioned.
• _section_rank replaced with an anchor-only ordering model: Start Here and Overview are pinned first; Design & Notes, Testing, CI/CD and Release, and Supporting Material are pinned last; all other sections (LLM-named domain sections like "Order Management" or "Authentication & User Management") are ordered by first-appearance in the LLM's proposal.
• _normalize_nav_section no longer aggressively remaps domain section names (Architecture → Core Workflows, Subsystems → Core Workflows, etc.); only safe universal aliases remain.
• PROPOSE_PROMPT now instructs the LLM to name nav sections after business domains rather than generic technical layers.