from __future__ import annotations import argparse import json import os import re import time import urllib.error import urllib.request from pathlib import Path DEFAULT_API_KEY_ENVS = ["MINIMAX_API_KEY", "MINIMAX_KEY", "MINIMAX_TOKEN", "HAILUO_API_KEY", "HAILUO_TOKEN"] DEFAULT_MODEL = "MiniMax-M2.7" def read(path: Path) -> str: return path.read_text(encoding="utf-8") def write(path: Path, text: str) -> None: path.parent.mkdir(parents=True, exist_ok=True) path.write_text(text.rstrip() + "\n", encoding="utf-8") def api_key() -> tuple[str, str]: for name in DEFAULT_API_KEY_ENVS: value = os.environ.get(name) if value: return name, value raise RuntimeError("Missing MiniMax API key") def endpoints() -> list[str]: bases = [] if os.environ.get("MINIMAX_BASE_URL"): bases.append(os.environ["MINIMAX_BASE_URL"]) bases.extend(["https://api.minimaxi.com", "https://api.minimax.io"]) out = [] for base in bases: base = base.rstrip("/") endpoint = base if base.endswith("/v1/text/chatcompletion_v2") else f"{base}/v1/text/chatcompletion_v2" if endpoint not in out: out.append(endpoint) return out def call_minimax(prompt: str, model: str, timeout: int, max_completion_tokens: int) -> tuple[str, dict, str]: key_name, key = api_key() payload = { "model": model, "messages": [ { "role": "system", "name": "MiniMax AI", "content": "You are a senior Chinese editor for game narrative, UX, and game-design books.", }, {"role": "user", "name": "user", "content": prompt}, ], "stream": False, "temperature": 0.35, "top_p": 0.9, "max_completion_tokens": max_completion_tokens, } body = json.dumps(payload, ensure_ascii=False).encode("utf-8") last_error: Exception | None = None for endpoint in endpoints(): for attempt in range(1, 4): try: request = urllib.request.Request( endpoint, data=body, headers={"Authorization": f"Bearer {key}", "Content-Type": "application/json"}, method="POST", ) with urllib.request.urlopen(request, timeout=timeout) as response: data = json.loads(response.read().decode("utf-8")) base_resp = data.get("base_resp") or {} if base_resp.get("status_code") not in (None, 0): raise RuntimeError(f"MiniMax base_resp error: {base_resp}") content = ((data.get("choices") or [{}])[0].get("message") or {}).get("content") or "" if not content.strip(): raise RuntimeError("MiniMax returned empty content") return content.strip(), {"api_key_env": key_name, "usage": data.get("usage"), "finish_reason": (data.get("choices") or [{}])[0].get("finish_reason")}, endpoint except (urllib.error.URLError, urllib.error.HTTPError, TimeoutError, RuntimeError) as exc: last_error = exc print(f" attempt {attempt} failed at {endpoint}: {type(exc).__name__}: {exc}") time.sleep(min(20, attempt * 3)) raise RuntimeError(f"MiniMax call failed: {last_error}") def split_markdown(text: str, max_chars: int) -> list[str]: paras = re.split(r"\n{2,}", text.strip()) chunks: list[str] = [] current = "" for para in paras: if current and len(current) + len(para) + 2 > max_chars: chunks.append(current.strip()) current = "" if len(para) > max_chars: lines = para.splitlines() for line in lines: if current and len(current) + len(line) + 1 > max_chars: chunks.append(current.strip()) current = "" current = f"{current}\n{line}".strip() if current else line else: current = f"{current}\n\n{para}".strip() if current else para if current: chunks.append(current.strip()) return chunks def polish_prompt(chunk: str, index: int, total: int) -> str: return f"""Polish the following Chinese Markdown translation of a game narrative/UX design book. Hard requirements: - Output polished Chinese Markdown only. - Do not summarize, omit, expand, or invent content. - Preserve all Markdown headings, tables, bullet structure, image links, figure numbers, source-page metadata, bibliography entries, URLs, and proper nouns. - Keep image Markdown exactly if present. - Improve Chinese fluency, remove translationese, clarify argument flow, and stabilize terminology. - Keep English proper nouns and book/game/theory names when appropriate. - Translate or smooth awkward mixed-language fragments unless they are proper nouns. - This is chunk {index}/{total}; polish only this chunk. Preferred terms: game narrative=游戏叙事; narrative design=叙事设计; UX design=UX 设计/用户体验设计; agency=主体能动性; affordance=可供性; signifier=信号; onboarding=新手引导; player journey=玩家旅程; emotion mapping=情绪映射; core loop=核心循环; world-building=世界观构建; engagement=参与度. Markdown chunk: {chunk} """ def polish_chapters(root: Path, model: str, timeout: int, chunk_chars: int, max_completion_tokens: int) -> list[dict]: src_dir = root / "chapters" out_dir = root / "chapters_polished" run_dir = root / "work" / "minimax_polish_runs" out_dir.mkdir(parents=True, exist_ok=True) run_dir.mkdir(parents=True, exist_ok=True) metas = [] for chapter in sorted(src_dir.glob("*.md")): text = read(chapter) chunks = split_markdown(text, chunk_chars) print(f"POLISH {chapter.name} chunks={len(chunks)} chars={len(text)}") polished_chunks = [] chunk_metas = [] for index, chunk in enumerate(chunks, start=1): print(f" chunk {index}/{len(chunks)} chars={len(chunk)}") content, meta, endpoint = call_minimax(polish_prompt(chunk, index, len(chunks)), model, timeout, max_completion_tokens) polished_chunks.append(content) chunk_metas.append({**meta, "endpoint": endpoint, "chunk_index": index, "chunk_total": len(chunks), "source_chars": len(chunk)}) time.sleep(0.8) output = "\n\n".join(part.strip() for part in polished_chunks if part.strip()) write(out_dir / chapter.name, output) meta = {"chapter": chapter.name, "chunks": chunk_metas} write(run_dir / f"{chapter.stem}-{time.strftime('%Y%m%d-%H%M%S')}.json", json.dumps(meta, ensure_ascii=False, indent=2)) metas.append(meta) return metas def good_sentence_prompt(items: list[tuple[str, str]]) -> str: joined = "\n\n".join(f"## {title}\n{text[:9000]}" for title, text in items) return f"""Create a Chinese reading-note document for memorable ideas from this game narrative/UX book. Copyright boundary: - Do not reproduce long English sentences. - Use at most 10 total English trigger phrases. - Each English trigger phrase must be 1-3 words only. - Most content must be Chinese paraphrase, not English quotation. Output Markdown in Chinese with: 1. A short copyright/use-boundary note. 2. A table with columns: 章节, 英文触发词, 中文好句摘要, 可借鉴的写法, 可用于哪类设计讨论. 3. 24-36 rows total. Source excerpts: {joined} """ def build_good_sentence_notes(root: Path, model: str, timeout: int, max_completion_tokens: int) -> None: outline = read(root / "source_outline.md") rows = [] for line in outline.splitlines(): if not line.startswith("| ") or "`work/extracted_text/" not in line: continue cells = [cell.strip() for cell in line.strip("|").split("|")] if len(cells) != 6 or cells[0] == "序号": continue title = cells[2] match = re.search(r"`work/extracted_text/([^`]+)`", cells[4]) if not match: continue source = root / "work" / "extracted_text" / match.group(1) rows.append((title, read(source))) # Keep source small enough for a single selection pass while covering all sections. content, meta, endpoint = call_minimax(good_sentence_prompt(rows), model, timeout, max_completion_tokens) out = root / "03-英文好句摘要与译法.md" write(out, content) write(root / "work" / "good_sentence_notes_run.json", json.dumps({**meta, "endpoint": endpoint}, ensure_ascii=False, indent=2)) def regenerate_combined(root: Path) -> None: out = root / "00-全书精校版.md" parts = [ "# Game Narrative Design and UX Fundamentals 中文精校版\n", "> 状态：MiniMax API 初译后进行中文审校润色，图片已按 PDF 裁切回填，仍建议人工逐章复核。 \n" "> 使用边界：个人学习 / 内部研究；公开传播或商业使用前需要自行确认授权。\n", ] for chapter in sorted((root / "chapters_polished").glob("*.md")): content = read(chapter).strip().replace("../assets/", "assets/") parts.append("\n\n---\n\n" + content) write(out, "\n".join(parts)) def verify(root: Path) -> str: assets = {p.name for p in (root / "assets").glob("fig-*.png")} polished = "\n".join(read(p) for p in (root / "chapters_polished").glob("*.md")) refs = set(re.findall(r"!\[[^\]]*\]\(\.\./assets/(fig-[^)]+\.png)\)", polished)) combined = read(root / "00-全书精校版.md") combined_refs = set(re.findall(r"!\[[^\]]*\]\(assets/(fig-[^)]+\.png)\)", combined)) forbidden = re.findall(r"pending|---TRANSLATION---|---NEW_TERMS---|作为 AI|作为AI|I am an AI|```", polished + "\n" + combined) lines = [ "# 质量门报告", "", "## 本轮状态", "", "- 文本状态：已生成 `chapters_polished/` 与 `00-全书精校版.md`。", "- 图片状态：精校版保留并回填全部 Figure 图片引用。", "- 英文好句：已生成 `03-英文好句摘要与译法.md`，英文仅保留短触发词，不做长段原文搬运。", "", "## 自动检查", "", f"- 图片资产数：{len(assets)}。", f"- 分章图片引用数：{len(refs)}。", f"- 合并版图片引用数：{len(combined_refs)}。", f"- 分章缺失图片引用：{', '.join(sorted(assets - refs)) if assets - refs else '无'}。", f"- 合并版缺失图片引用：{', '.join(sorted(assets - combined_refs)) if assets - combined_refs else '无'}。", f"- 模型分隔符/AI 自称/待翻译占位残留：{len(forbidden)}。", "", "## 仍需人工判断", "", "- 图内英文标签尚未逐张中文化重绘；目前是原图裁切回填。", "- 英文好句为短摘和中文摘要，不构成完整原文摘抄。", "- 精校版经过模型润色，但公开传播前仍需人工审校和授权确认。", ] report = "\n".join(lines) write(root / "audit" / "质量门报告.md", report) return report def update_readme(root: Path) -> None: readme = root / "README.md" text = read(readme) text = re.sub( r"状态：.*", "状态：MiniMax API 初译完成；中文审校润色版已生成；已完成 12/12 个翻译单元；已回填 46 张 Figure 裁切图 ", text, count=1, ) additions = [ "- `00-全书精校版.md`：中文审校润色后的合并通读版。", "- `chapters_polished/`：中文审校润色后的分章版本。", "- `03-英文好句摘要与译法.md`：英文短触发词 + 中文好句摘要 + 译法点评。", "- `audit/质量门报告.md`：本轮自动检查与人工复核边界。", ] for line in reversed(additions): if line not in text: text = text.replace("## 文件说明\n\n", f"## 文件说明\n\n{line}\n") write(readme, text) def main() -> None: parser = argparse.ArgumentParser() parser.add_argument("--root", required=True) parser.add_argument("--model", default=os.environ.get("MINIMAX_MODEL", DEFAULT_MODEL)) parser.add_argument("--timeout", type=int, default=420) parser.add_argument("--chunk-chars", type=int, default=18000) parser.add_argument("--max-completion-tokens", type=int, default=26000) parser.add_argument("--skip-polish", action="store_true") args = parser.parse_args() root = Path(args.root) if not args.skip_polish: polish_chapters(root, args.model, args.timeout, args.chunk_chars, args.max_completion_tokens) build_good_sentence_notes(root, args.model, args.timeout, args.max_completion_tokens) regenerate_combined(root) report = verify(root) update_readme(root) print(report) if __name__ == "__main__": main()