# Multi-Version Contract Comparison Table (三版对比表)

## When to Use
When Maggie/Doro asks to compare multiple versions of a contract (typically: template / counterparty revision / our revision), produce a structured docx comparison table.

## Pattern (2026-07-03 模特合作协议 session)

### Document Setup
- **Landscape orientation** for 4-5 columns: `section.orientation = 1; page_width=Cm(29.7); page_height=Cm(21.0)`
- Narrow margins: 1.2-1.5cm all sides
- Font size 8.5-9pt for table cells (fits more content)

### Table Structure
| 条款 | 【模版】 | 版本A（对方修订） | 版本B（我方修订） | 双方协商一致 |
|------|---------|------------------|-----------------|-------------|

### Red Font for Differences
- Column N is red when its content differs from other versions
- Use `RGBColor(0xFF, 0x00, 0x00)` on the run
- "协商一致" column: red = current text doesn't match consensus → needs modification

### Yellow Background for Consensus Column
```python
def set_cell_shading(cell, color):
    tc = cell._element
    tcPr = tc.find(qn('w:tcPr'))
    if tcPr is None:
        tcPr = OxmlElement('w:tcPr')
        tc.insert(0, tcPr)
    shading = OxmlElement('w:shd')
    shading.set(qn('w:fill'), color)  # e.g. 'FFF8E1' for light yellow
    shading.set(qn('w:val'), 'clear')
    tcPr.append(shading)
```

### Header Row Styling
- Blue background (`D9E2F3`)
- Bold, centered, font size 8.5pt

### Data Structure in Code
```python
# Each row: (clause_name, col1_text, col2_text, col3_text, col4_text, col2_red, col3_red, col4_red)
rows = [
    ('条款名',
     '模版内容',
     '对方修订内容',
     '我方修订内容', 
     '协商一致内容',
     True,   # col2 red? (differs from others)
     False,  # col3 red?
     True),  # col4 red? (doesn't match consensus)
]
```

### Legend at Bottom
Include a legend explaining what red means in each column:
- 版本A列红色 = 与模版/我方版不一致（对方的修改）
- 版本B列红色 = 与模版不一致（我方的修改）
- 协商一致列红色 = 当前文本与协商一致不符，需要修改

## Key Lessons
1. **Read all three files from Nextcloud** using `sudo find ~/nextcloud/data/data/...` path
2. **Extract paragraph text** using python-docx: `[(i, p.text.strip()) for i, p in enumerate(doc.paragraphs) if p.text.strip()]`
3. **Check tables** separately: `doc.tables` — contracts often have signature blocks and SNS account tables
4. **Align comparison by clause semantics**, not paragraph index — different versions may have different paragraph counts
5. Also upload to Nextcloud for viewing in OnlyOffice

## Per-Run Precision for Tracked Changes (Maggie's correction)

When applying tracked changes based on comparison results, **never replace entire paragraphs**. Instead:

1. Identify the specific runs containing text to change
2. For each run: create w:del wrapping a deepcopy (converting w:t → w:delText), create w:ins with new text and cloned rPr, swap in place
3. All surrounding runs remain untouched

```python
# Find specific run by text content
for r in para_element.findall(qn('w:r')):
    text = ''.join(t.text or '' for t in r.findall(qn('w:t')))
    if text == '¥700,000':  # exact match on this run
        r_parent = r.getparent()
        r_idx = list(r_parent).index(r)
        # Create del wrapping copy of this run
        del_elem = make_del_run_from_existing(r)
        # Create ins with new value, same rPr
        ins_elem = make_ins_run('¥600,000', r.find(qn('w:rPr')))
        r_parent.remove(r)
        r_parent.insert(r_idx, ins_elem)
        r_parent.insert(r_idx, del_elem)
        break
```

This produces clean tracked changes where Word/OnlyOffice shows exactly which characters changed (e.g., ~~700,000~~ → 600,000) rather than entire-paragraph replacements.