WB INS Font Fix + Bold-Stripping Disaster (2026-06-10)

Problem

Doro rejected 【修】青浦精神卫生-委托检验协议（修）.docx: WB insertions displayed in wrong font.

Terminal review passed 10 checks but missed basic font inconsistency.

Root Cause

contract_docx_lib.py's _ensure_rfonts_complete() only filled hAnsi/cs, missing:

w:eastAsia — CJK falls back to docDefaults (Times New Roman vs 宋体)
w:hint="eastAsia" — required for CJK rendering priority
w:sz — not inherited inside w:ins; must be explicit

Fix Applied (contract_docx_lib.py, 4 locations)

1. `_ensure_rfonts_complete()` — fill all 4 + hint

2. `_extract_formats()` — ensure sz on body_rpr/title_rpr

3. `tracked_replace()` — ensure sz on INS rpr

4. `tracked_replace()` — fallback for hint-only rFonts runs

See previous version of this file for code details.

CRITICAL: Bold-Stripping Disaster

What happened

First fix attempt: after noticing font issues, batch-stripped ALL <w:b/> tags from all WB INS runs. This destroyed the original bold formatting:

P19: 风险律师费条件（原文加粗）→ stripped → not bold ✗
P18: 金额部分（原文加粗）→ stripped → not bold ✗
P30: "如若违反，律师费将另行重新计付"（原文加粗）→ stripped → not bold ✗

Doro: "要保持原文的格式不变"

Why it was wrong

tracked_replace correctly inherits the original run's rPr INCLUDING bold. When the original text "在代理过程中..." was bold, the replacement text "风险律师费仅按..." SHOULD also be bold — that's correct format preservation.

The rule

tracked_replace INS inherits original run's rPr → leave it alone
add_clause uses _body_rpr (no bold) or _title_rpr (bold) → correct by design
NEVER post-process WB INS to strip/add formatting attributes globally

Correct validation approach

Compare each WB INS run against the original run in the same paragraph, not against a global template. Use scripts/wb-ins-font-verify.py for document-agnostic validation.

Verification Script

scripts/wb-ins-font-verify.py <docx> — per-paragraph comparison, checks rFonts/sz/hint, exit code 0=pass 1=fail. Document-agnostic (no hardcoded font names).

Three font-source scenarios in tracked_replace

Run has explicit font names → _ensure_rfonts_complete fills gaps
Run has hint-only rFonts (no names) → fallback to _body_rpr font names
Run has no rPr at all → use _body_rpr directly

2.5 KiB Raw Permalink Blame History