Closes RFC planning/mem-add-tags-blockquote-promote-rfc.md §Follow-ups #2. Independent of #465 (broadcast cleanup) — different files.

Summary

When chunk_file strips a section-leading > created: / > tags: blockquote before passing text to _split_section, base_line still points at the heading line. line_offset started at 0 and counted only body lines, so sub-chunks 2..N reported start_line 3–5 lines earlier than the body they actually covered.

Practical impact: mem_edit on a non-first sub-chunk reads file_lines[start_line..end_line] via replace_chunk_body. With drift, those lines included real body content from the previous sub-chunk — which _find_body_start_index does not classify as a header (no heading, no leading blockquote at the range start), so preserved == [] and the replacement silently dropped that content on save.

What changed

• _extract_section_blockquote_tags returns a third value lines_consumed (count of input lines the strip consumed). Backwards-incompatible at the helper API surface, but it's a private classmethod with a single in-tree caller.
• chunk_file computes body_offset = 1 (heading) + leading_strip_lines + blockquote_strip_lines and passes it to _split_section as a kwarg. leading_strip_lines is the count of \n consumed by the existing .strip() on raw section text.
• _split_section accepts body_offset (default 0) and seeds line_offset with it. Default 0 = pre-strip behavior preserved for callers that don't strip.

Sub-chunk 1's start_line still anchors at the heading (preserves mem_edit header-preservation for the first sub-chunk; current_start = base_line at init is only consulted at the first emit boundary, by which time line_offset has accumulated past the seed).

Test plan

• New tests/test_chunking_blockquote_tags.py::TestOversizeSectionLineDrift:
  • test_blockquote_header_does_not_drag_sub_chunk_start_lines — controlled oversize section with explicit > created: / > tags: header; sub-chunks 2..N must report start_line >= body-start-file-line. Fails pre-fix.
  • test_oversize_section_without_blockquote_unchanged — no header to strip → body_offset = 0 → pre-strip behaviour preserved.
• Full suite: uv run pytest -m "not ollama" → 2405 passed, 0 failed (PR feat(mem_add,mem_edit): canonical writer + mem_edit header preservation #464 baseline 2403 + 2 new).
• uv run ruff check && uv run ruff format --check clean.
• uv run mypy packages/memtomem/src/memtomem/chunking/markdown.py — clean.

RFC §Follow-ups status after this PR

• ✅ chore: prepare for open-source release #1 — mem_edit header preservation (PR feat(mem_add,mem_edit): canonical writer + mem_edit header preservation #464)
• ✅ refactor(stm): decouple surfacing via remote-only MCP #2 — _split_section line drift (this PR)
• ✅ chore(core): strip STM integration residue #3 — block-list shape test (PR feat(mem_add,mem_edit): canonical writer + mem_edit header preservation #464)
• ⏳ chore: drop packages/memtomem-stm from monorepo (Phase 3b) #4 — add_entries_batch broadcast cleanup (PR fix(mem_batch_add): drop post-index tag broadcast; per-entry tags isolated #465, in CI)

After this + #465 merge, the only remaining item from the blockquote-tags RFC sequence is the chunk_links table — separate RFC.

🤖 Generated with Claude Code