inkos/packages/core/src/agents/post-write-validator.ts

/**
 * Post-write rule-based validator.
 *
 * Deterministic, zero-LLM-cost checks that run after every chapter generation.
 * Catches violations that prompt-only rules cannot guarantee.
 */

import type { BookRules } from "../models/book-rules.js";
import type { GenreProfile } from "../models/genre-profile.js";

export interface PostWriteViolation {
  readonly rule: string;
  readonly severity: "error" | "warning";
  readonly description: string;
  readonly suggestion: string;
}

// --- Marker word lists ---

/** AI转折/惊讶标记词 */
const SURPRISE_MARKERS = ["仿佛", "忽然", "竟然", "猛地", "猛然", "不禁", "宛如"];

/** 元叙事/编剧旁白模式 */
const META_NARRATION_PATTERNS = [
  /到这里[，,]?算是/,
  /接下来[，,]?(?:就是|将会|即将)/,
  /(?:后面|之后)[，,]?(?:会|将|还会)/,
  /(?:故事|剧情)(?:发展)?到了/,
  /读者[，,]?(?:可能|应该|也许)/,
  /我们[，,]?(?:可以|不妨|来看)/,
];

/** 分析报告式术语（禁止出现在正文中） */
const REPORT_TERMS = [
  "核心动机", "信息边界", "信息落差", "核心风险", "利益最大化",
  "当前处境", "行为约束", "性格过滤", "情绪外化", "锚定效应",
  "沉没成本", "认知共鸣",
];

/** 作者说教词 */
const SERMON_WORDS = ["显然", "毋庸置疑", "不言而喻", "众所周知", "不难看出"];

/** 全场震惊类集体反应 */
const COLLECTIVE_SHOCK_PATTERNS = [
  /(?:全场|众人|所有人|在场的人)[，,]?(?:都|全|齐齐|纷纷)?(?:震惊|惊呆|倒吸凉气|目瞪口呆|哗然|惊呼)/,
  /(?:全场|一片)[，,]?(?:寂静|哗然|沸腾|震动)/,
];

// --- Validator ---

export function validatePostWrite(
  content: string,
  genreProfile: GenreProfile,
  bookRules: BookRules | null,
): ReadonlyArray<PostWriteViolation> {
  const violations: PostWriteViolation[] = [];

  // Skip Chinese-specific rules for English content
  const isEnglish = genreProfile.language === "en";
  if (isEnglish) {
    // For English, only run book-specific prohibitions and paragraph length check
    return validatePostWriteEnglish(content, genreProfile, bookRules);
  }

  // 1. 硬性禁令: "不是…而是…" 句式
  if (/不是[^，。！？\n]{0,30}[，,]?\s*而是/.test(content)) {
    violations.push({
      rule: "禁止句式",
      severity: "error",
      description: "出现了「不是……而是……」句式",
      suggestion: "改用直述句",
    });
  }

  // 2. 硬性禁令: 破折号
  if (content.includes("——")) {
    violations.push({
      rule: "禁止破折号",
      severity: "error",
      description: "出现了破折号「——」",
      suggestion: "用逗号或句号断句",
    });
  }

  // 3. 转折/惊讶标记词密度 ≤ 1次/3000字
  const markerCounts: Record<string, number> = {};
  let totalMarkerCount = 0;
  for (const word of SURPRISE_MARKERS) {
    const matches = content.match(new RegExp(word, "g"));
    const count = matches?.length ?? 0;
    if (count > 0) {
      markerCounts[word] = count;
      totalMarkerCount += count;
    }
  }
  const markerLimit = Math.max(1, Math.floor(content.length / 3000));
  if (totalMarkerCount > markerLimit) {
    const detail = Object.entries(markerCounts)
      .map(([w, c]) => `"${w}"×${c}`)
      .join("、");
    violations.push({
      rule: "转折词密度",
      severity: "warning",
      description: `转折/惊讶标记词共${totalMarkerCount}次（上限${markerLimit}次/${content.length}字），明细：${detail}`,
      suggestion: "改用具体动作或感官描写传递突然性",
    });
  }

  // 4. 高疲劳词检查（从 genreProfile 读取，单章每词 ≤ 1次）
  const fatigueWords = bookRules?.fatigueWordsOverride && bookRules.fatigueWordsOverride.length > 0
    ? bookRules.fatigueWordsOverride
    : genreProfile.fatigueWords;
  for (const word of fatigueWords) {
    const escaped = word.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
    const matches = content.match(new RegExp(escaped, "g"));
    const count = matches?.length ?? 0;
    if (count > 1) {
      violations.push({
        rule: "高疲劳词",
        severity: "warning",
        description: `高疲劳词"${word}"出现${count}次（上限1次/章）`,
        suggestion: `替换多余的"${word}"为同义但不同形式的表达`,
      });
    }
  }

  // 5. 元叙事检查（编剧旁白）
  for (const pattern of META_NARRATION_PATTERNS) {
    const match = content.match(pattern);
    if (match) {
      violations.push({
        rule: "元叙事",
        severity: "warning",
        description: `出现编剧旁白式表述："${match[0]}"`,
        suggestion: "删除元叙事，让剧情自然展开",
      });
      break; // 报一次即可
    }
  }

  // 6. 分析报告式术语
  const foundTerms: string[] = [];
  for (const term of REPORT_TERMS) {
    if (content.includes(term)) {
      foundTerms.push(term);
    }
  }
  if (foundTerms.length > 0) {
    violations.push({
      rule: "报告术语",
      severity: "error",
      description: `正文中出现分析报告术语：${foundTerms.map(t => `"${t}"`).join("、")}`,
      suggestion: "这些术语只能用于 PRE_WRITE_CHECK 内部推理，正文中用口语化表达替代",
    });
  }

  // 7. 作者说教词
  const foundSermons: string[] = [];
  for (const word of SERMON_WORDS) {
    if (content.includes(word)) {
      foundSermons.push(word);
    }
  }
  if (foundSermons.length > 0) {
    violations.push({
      rule: "作者说教",
      severity: "warning",
      description: `出现说教词：${foundSermons.map(w => `"${w}"`).join("、")}`,
      suggestion: "删除说教词，让读者自己从情节中判断",
    });
  }

  // 8. 全场震惊类集体反应
  for (const pattern of COLLECTIVE_SHOCK_PATTERNS) {
    const match = content.match(pattern);
    if (match) {
      violations.push({
        rule: "集体反应",
        severity: "warning",
        description: `出现集体反应套话："${match[0]}"`,
        suggestion: "改写成1-2个具体角色的身体反应",
      });
      break;
    }
  }

  // 9. 连续"了"字检查（3句以上连续含"了"）
  const sentences = content
    .split(/[。！？]/)
    .map(s => s.trim())
    .filter(s => s.length > 2);

  let consecutiveLe = 0;
  let maxConsecutiveLe = 0;
  for (const sentence of sentences) {
    if (sentence.includes("了")) {
      consecutiveLe++;
      maxConsecutiveLe = Math.max(maxConsecutiveLe, consecutiveLe);
    } else {
      consecutiveLe = 0;
    }
  }
  if (maxConsecutiveLe >= 6) {
    violations.push({
      rule: "连续了字",
      severity: "warning",
      description: `检测到${maxConsecutiveLe}句连续包含"了"字，节奏拖沓`,
      suggestion: "保留最有力的一个「了」，其余改为无「了」句式",
    });
  }

  // 10. 段落长度检查（手机阅读适配：50-250字/段为宜）
  const paragraphs = content
    .split(/\n\s*\n/)
    .map(p => p.trim())
    .filter(p => p.length > 0);

  const longParagraphs = paragraphs.filter(p => p.length > 300);
  if (longParagraphs.length >= 2) {
    violations.push({
      rule: "段落过长",
      severity: "warning",
      description: `${longParagraphs.length}个段落超过300字，不适合手机阅读`,
      suggestion: "长段落拆分为3-5行的短段落，在动作切换或情绪节点处断开",
    });
  }

  // 11. Book-level prohibitions
  // Short prohibitions (2-30 chars): exact substring match
  // Long prohibitions (>30 chars): skip — these are conceptual rules for prompt-level enforcement only
  if (bookRules?.prohibitions) {
    for (const prohibition of bookRules.prohibitions) {
      if (prohibition.length >= 2 && prohibition.length <= 30 && content.includes(prohibition)) {
        violations.push({
          rule: "本书禁忌",
          severity: "error",
          description: `出现了本书禁忌内容："${prohibition}"`,
          suggestion: "删除或改写该内容",
        });
      }
    }
  }

  return violations;
}

/** English-specific post-write validation rules. */
function validatePostWriteEnglish(
  content: string,
  genreProfile: GenreProfile,
  bookRules: BookRules | null,
): ReadonlyArray<PostWriteViolation> {
  const violations: PostWriteViolation[] = [];

  // 1. AI-tell word density (from en-prompt-sections IRON LAW 3)
  const aiTellWords = ["delve", "tapestry", "testament", "intricate", "pivotal", "vibrant", "embark", "comprehensive", "nuanced"];
  for (const word of aiTellWords) {
    const regex = new RegExp(`\\b${word}\\b`, "gi");
    const matches = content.match(regex);
    if (matches && matches.length > Math.ceil(content.length / 3000)) {
      violations.push({
        rule: "AI-tell word density",
        severity: "warning",
        description: `"${word}" appears ${matches.length} times (limit: 1 per 3000 chars)`,
        suggestion: `Replace with a more specific word`,
      });
    }
  }

  // 2. Paragraph overflow (same rule applies to English)
  const paragraphs = content.split(/\n\s*\n/).filter((p) => p.trim().length > 0);
  const longParagraphs = paragraphs.filter((p) => p.length > 500);
  if (longParagraphs.length >= 2) {
    violations.push({
      rule: "Paragraph length",
      severity: "warning",
      description: `${longParagraphs.length} paragraphs exceed 500 characters`,
      suggestion: "Break into shorter paragraphs for readability",
    });
  }

  // 3. Book-specific prohibitions
  if (bookRules?.prohibitions) {
    for (const prohibition of bookRules.prohibitions) {
      if (prohibition.length >= 2 && prohibition.length <= 50 && content.toLowerCase().includes(prohibition.toLowerCase())) {
        violations.push({
          rule: "Book prohibition",
          severity: "error",
          description: `Found banned content: "${prohibition}"`,
          suggestion: "Remove or rewrite this content",
        });
      }
    }
  }

  // 4. Genre fatigue words
  const fatigueWords = bookRules?.fatigueWordsOverride && bookRules.fatigueWordsOverride.length > 0
    ? bookRules.fatigueWordsOverride
    : genreProfile.fatigueWords;
  for (const word of fatigueWords) {
    const regex = new RegExp(`\\b${word}\\b`, "gi");
    const matches = content.match(regex);
    if (matches && matches.length > 1) {
      violations.push({
        rule: "Fatigue word",
        severity: "warning",
        description: `"${word}" appears ${matches.length} times (max 1 per chapter)`,
        suggestion: "Vary the vocabulary",
      });
    }
  }

  return violations;
}