v0.4.3

Fixed

• Reprinted continuation-page headers appended as data rows on multi-page
  merge (adapters/docling.py). When a table's column header is reprinted at
  the top of each page — especially a multi-row (hierarchical) header — the
  repeated header rows survived the merge as bogus data rows, misaligning the
  stitched table. Injection now drops a body row when it is both flagged
  column_header by Docling and a tokenized match (Jaccard ≥ 0.6) for the
  reconstructed header block. Both signals are required: the flag alone is
  unreliable (Docling over-flags rowspan/continuation data rows as headers),
  and the tokenized comparison is punctuation-agnostic, so per-cell OCR drift
  such as (S$) vs ($$) is tolerated without any threshold tuning. The merged
  DataFrame (lt.df) is unchanged; only the injected document is de-duplicated.
  A debug log reports each dropped row.

v0.4.2

Fixed

• __version__ was hardcoded and stale (__init__.py). It read "0.2.0"
  regardless of the installed release, since nothing tied it to the version in
  pyproject.toml. It is now derived from the installed distribution metadata
  via importlib.metadata.version("table-stitcher"), so it always reflects the
  actual release (falling back to "0.0.0+unknown" when run from an
  uninstalled source tree). The release gate now also asserts
  __version__ matches the pyproject.toml version, so the two can't drift
  again.

v0.4.1

Fixed

• Spanning body cells duplicated across columns on multi-page merge
  (adapters/docling.py). Docling repeats a col_span=N cell's text across
  every column it covers; the merge round-trip rebuilt those as N separate
  col_span=1 cells, leaking a full-width description into every value column
  and displacing the real values (a repeated col_span header behaved the same
  way). Injection now matches each merged row back to its source grid row and
  re-emits the original spans; rows the merger transformed (stitched
  continuations, folded overflow) fall back to the flat 1x1 rebuild. The match
  uses the original span metadata, never value equality, so coincidentally-equal
  adjacent values (e.g. two plan columns sharing a cap) stay separate cells.

v0.4.0

Added — intervening-content guard (#10)

Two tables that share a column schema but belong to different sections (a heading sits between them in reading order) are no longer stitched into one. A genuine page-split continuation has only page furniture between its fragments, so a section heading between two tables is a reliable "separate tables" signal.

• New MultiPageConfig.block_on_intervening_content (default True) and TableMeta.content_before; both merge paths (_classify_sequential_pair and should_force_orphan_merge) consult it.
• Running headers — including a banner one page labels page_header and another mislabels section_header — are detected via near-identical (Jaccard >= 0.8) cross-page recurrence, so legitimate continuations still merge.
• Only section_header/title block; paragraphs, list items, captions, footnotes and figures are ignored.

Fixes over-eager merging of same-schema per-section tables (e.g. an insurance policy's eight Prestige | Elite | Classic benefit grids collapsing into one).

v0.3.0

Fixed

• Category rows incorrectly folded into preceding data rows (merger.py).
  stitch_split_cells() previously folded any row with exactly one non-empty
  cell into the row above it. Category/section-header rows (e.g. "Theme 2:
  Trust and Credibility" with text only in col 0) matched this pattern and
  were silently merged into the preceding data row, mangling participant IDs
  and destroying table structure. The fix: a non-empty col 0 in the candidate
  row signals a new record or section header — not an overflow — and folding
  is skipped. Legitimate split-cell continuations always have col 0 empty.
  Six existing fixture YAMLs updated to reflect the corrected (higher) row
  counts — the old YAMLs encoded the buggy folded output.

• False merge of independent same-width headerless tables (merger.py).
  When two adjacent tables both have is_headerless=True and the same column
  count, the merger now requires a layout signal (the left table must end near
  the bottom of its page, vert_bottom >= bottom_band_min) before merging.
  Previously, column count alone was sufficient — three independent clinical
  lab panels (each 4 columns, no header row) collapsed into one 22-row table.
  Legitimate multi-page headerless tables are unaffected: they fill their pages
  and always produce a strong layout signal.

Added

• Parser-neutral YAML fixture layer (tests/fixtures/tablemeta/) plus
  tests/test_tablemeta_fixtures.py. New adapters can validate against the
  merger's full test surface by feeding the same YAMLs through their own
  extract() — no PDF or OCR involvement.
• Public-API integration coverage: every fixture now runs through both
  merge_multipage_tables() (parser-neutral) and stitch_tables()
  (full pipeline including docling injection).
• scripts/release_gate.sh — offline-friendly release gate that runs unit
  tests, rebuilds dist/, installs the wheel into a clean venv, and
  smoke-tests the installed package. RELEASE_GATE_ONLINE=1 toggles
  isolated build/install for CI.

Changed

• Core merger refactored for readability. merge_multipage_tables() is
  now a four-phase orchestrator (setup → pass 1 sequential → pass 2 orphan repair → build) delegating to named helpers. _classify_sequential_pair()
  isolates adjacent-pair merge logic for independent review. Behavior is
  unchanged — 127 tests prove equivalence.
• align_dataframe_to_header() dispatches to per-policy handlers
  (_overflow_preserve_extra, _overflow_warn_drop, _overflow_fail,
  _overflow_merge_tail) instead of branching inline.

Removed

• Dead pos_to_orig variable in the merger setup path.