feat(evaluation): memoir readiness, judge/replay updates, eval web playground

Add memoir_readiness_service and router tests; extend judge schemas/services, replay_service, and conversation rubric; align story route agent, payload, prompts, and story_pipeline_sync; update agent logging, config, and DI. Document internal-eval; add replayDraft util and PlaygroundPage changes in app-eval-web.

api/app/agents/memoir/prompts.py

@@ -425,8 +425,9 @@ def story_route_merge_hint_for_category(chapter_category: str) -> str:
     if cc == "family":
         return (
             "### 本章类别路由倾向（家庭）\n"
-            "- 原则性反思、关系模式、相处之道的补充 → **倾向 append_story**。\n"
-            "- **明确的新事件链**（新场景、新时间线、不同人物组合的新经历）→ 可 new_story。"
+            "- **默认 append_story**：同一家庭成员、同一居住环境、婚姻育儿、节日团聚、童年与父母的回忆等，"
+            "只要仍围绕已出现的人物或关系网络补充细节，一律并入最匹配的候选，不要因为换了个场景就 new_story。\n"
+            "- 仅当口述出现**完全新的人物组合 + 可独立成篇的新事件链**（与所有候选正文都接不上）时，才 new_story。"
         )
     if cc in (
         "childhood",
@@ -435,6 +436,13 @@ def story_route_merge_hint_for_category(chapter_category: str) -> str:
         "career_achievement",
         "career_challenge",
     ):
+        if cc in ("childhood", "education"):
+            return (
+                "### 本章类别路由倾向（童年 / 求学 — 少拆分）\n"
+                "- **默认 append_story**：同一成长阶段里，地点（老家、学校）、父母职业、玩伴、游戏影视、"
+                "怀旧细节等**主题延续**的补充，即使分段讲述，也应并入已有童年/求学故事，避免多篇开头重复交代背景。\n"
+                "- **仅当**口述出现**另一条清晰可辨的事件链**（时间/地点/人物线换了且与候选明显不是同一脉络）时，才 new_story。"
+            )
         return (
             "### 本章类别路由倾向（经历叙事）\n"
             "- 以具体事件链为主：**不同事件 / 时期 / 地点** → 可 new_story。\n"
@@ -473,7 +481,7 @@ def get_story_route_prompt(
 
 **路由边界（必须遵守）**：仅根据下方「本批口述合并文本」判断；不得将系统检索摘要、记忆摘录等当作本批口述内容来匹配候选。
 
-**候选故事说明**：列表项可能含 `summary` 或 `body_for_route`（正文摘要）；仅含 `preview` 者为索引项，信息不全。**append 时优先匹配带 summary 或 body 的条目**；索引项仅作候选 id 备忘。
+**候选故事说明**：列表项可能含 `summary`、`body_for_route`（正文摘要）或 `opening_snippet`（无 summary 时的纯文本开头提要）；仅含 `preview` 者为索引项，信息不全。**append 时优先匹配带 summary / body / opening_snippet 的条目**；索引项仅作候选 id 备忘。
 
 当前章节（写作容器）：
 - category: {chapter_category}
@@ -495,6 +503,7 @@ def get_story_route_prompt(
 规则：
 - **不要**只因「不太确定」就选 new_story；在主题可并入某一候选时应 append_story。
 - 仅当口述与**所有**候选在两层标准下都明显不兼容时，才选 new_story。
+- 若已有候选故事（列表非空）且口述是对同一人生阶段的**补述**，却找不到精确 id：仍应 **append_story** 到最相近的一条，而不是 new_story。
 """
 
 
@@ -525,7 +534,7 @@ def get_story_batch_plan_prompt(
 - **append_story**：与某一候选在两层标准下可合并，且能对应到具体 candidate id
 - **new_story**：该块与**所有**候选都明显不兼容，或确认为独立新经历
 
-**候选故事说明**：条目可能含 `summary`/`body_for_route`；仅 `preview` 者为索引项。**优先用带摘要/正文的条目做 append 目标**。
+**候选故事说明**：条目可能含 `summary` / `body_for_route` / `opening_snippet`；仅 `preview` 者为索引项。**优先用带摘要、正文摘要或开头提要的条目做 append 目标**。
 
 当前章节（写作容器）：
 - category: {chapter_category}
@@ -552,6 +561,8 @@ def get_story_batch_plan_prompt(
 规则：
 - `units` 中所有 `segment_ids` 拼接后，必须**不重不漏**地覆盖本批全部 id，且顺序与【本批口述片段】数组一致
 - **不要**仅因不确定就对整块选 new_story；能并入候选时应 append_story
+- **同一批里 new_story 单元至多 1 个**：除非口述中同时存在**至少两条**与所有候选都不兼容、且彼此也明显无关的独立长经历，否则禁止拆成多个 new_story；连续多段若都在补充同一主题，应合并为**一块 append_story**。
+- 候选列表非空时，优先把本批当作「加厚已有篇章」，而不是再开新篇。
 """
 
 

api/app/agents/memoir/story_route_agent.py

@@ -13,7 +13,10 @@ from app.agents.memoir.prompts import (
     get_story_batch_plan_prompt,
     get_story_route_prompt,
 )
-from app.agents.memoir.story_route_payload import build_route_candidate_json
+from app.agents.memoir.story_route_payload import (
+    build_route_candidate_json,
+    sort_stories_for_route,
+)
 from app.core.config import settings
 from app.core.llm_call import LLMCallError, llm_json_call
 from app.core.logging import get_logger
@@ -25,6 +28,105 @@ logger = get_logger(__name__)
 # 超过此数量跳过批量规划（单次路由），避免 prompt 过大
 PLAN_BATCH_MAX_SEGMENTS = 48
 
+# 童年 / 求学 / 家庭：模型与后处理均倾向「少拆分、优先续写」
+APPEND_FIRST_CHAPTER_CATEGORIES = frozenset({"childhood", "education", "family"})
+
+
+def default_append_target_story_id(
+    candidate_stories: list[Story],
+    story_meta: dict[str, dict[str, int]] | None,
+    settings: Any,
+) -> str | None:
+    """排序后的首选续写目标（与路由候选 JSON 顺序一致）。"""
+    if not candidate_stories:
+        return None
+    meta = story_meta or {}
+    ordered = sort_stories_for_route(
+        candidate_stories,
+        meta,
+        summary_min_chars=int(settings.story_route_summary_min_chars),
+    )
+    if not ordered:
+        return None
+    return str(ordered[0].id)
+
+
+def merge_consecutive_new_story_units(
+    units: list[StoryBatchPlanUnit],
+) -> list[StoryBatchPlanUnit]:
+    """将相邻的多个 new_story 单元合并为一个，减少同批碎片叙事。"""
+    if not units:
+        return units
+    out: list[StoryBatchPlanUnit] = []
+    i = 0
+    while i < len(units):
+        u = units[i]
+        if u.decision != "new_story":
+            out.append(u)
+            i += 1
+            continue
+        run_segs: list[str] = list(u.segment_ids)
+        j = i + 1
+        while j < len(units) and units[j].decision == "new_story":
+            run_segs.extend(units[j].segment_ids)
+            j += 1
+        if j > i + 1:
+            out.append(
+                StoryBatchPlanUnit(
+                    segment_ids=run_segs,
+                    decision="new_story",
+                    target_story_id=None,
+                    reason="coalesced_consecutive_new_story",
+                )
+            )
+        else:
+            out.append(u)
+        i = j
+    return out
+
+
+def normalize_batch_plan_reduce_new_story_fragmentation(
+    plan: StoryBatchPlan,
+    ordered_segment_ids: list[str],
+    *,
+    chapter_category: str,
+    candidate_stories: list[Story],
+    valid_story_ids: set[str],
+    story_meta: dict[str, dict[str, int]] | None,
+    settings: Any,
+) -> StoryBatchPlan:
+    """
+    LLM 校验通过后的确定性归一：合并相邻 new_story；在 append-first 类目下若整批只有一个 new 块则改为 append。
+    """
+    units = merge_consecutive_new_story_units(list(plan.units))
+    if (
+        chapter_category in APPEND_FIRST_CHAPTER_CATEGORIES
+        and candidate_stories
+        and len(units) == 1
+        and units[0].decision == "new_story"
+    ):
+        tid = default_append_target_story_id(candidate_stories, story_meta, settings)
+        if tid and tid in valid_story_ids:
+            units = [
+                StoryBatchPlanUnit(
+                    segment_ids=list(ordered_segment_ids),
+                    decision="append_story",
+                    target_story_id=tid,
+                    reason="append_first_whole_batch_fallback",
+                )
+            ]
+    candidate = StoryBatchPlan(units=units)
+    ok, err = validate_story_batch_plan(
+        ordered_segment_ids, candidate, valid_story_ids
+    )
+    if not ok:
+        logger.warning(
+            "batch_plan_normalize_revalidate_failed err={} keep_original",
+            err,
+        )
+        return plan
+    return candidate
+
 
 class StoryBatchPlanUnit(BaseModel):
     """批量写入中的一个单元（连续 segment 块）。"""
@@ -120,6 +222,15 @@ class StoryRouteAgent:
         story_meta: dict[str, dict[str, int]] | None = None,
     ) -> StoryRouteDecision:
         if not llm:
+            fb = default_append_target_story_id(
+                candidate_stories, story_meta, settings
+            )
+            if fb and fb in valid_story_ids:
+                return StoryRouteDecision(
+                    decision="append_story",
+                    target_story_id=fb,
+                    reason="no_llm_default_append",
+                )
             return StoryRouteDecision(
                 decision="new_story",
                 new_story_title=None,
@@ -134,6 +245,15 @@ class StoryRouteAgent:
         )
 
         def _decide_fallback() -> StoryRouteDecision:
+            fb = default_append_target_story_id(
+                candidate_stories, story_meta, settings
+            )
+            if fb and fb in valid_story_ids:
+                return StoryRouteDecision(
+                    decision="append_story",
+                    target_story_id=fb,
+                    reason="parse_error_default_append",
+                )
             return StoryRouteDecision(
                 decision="new_story",
                 new_story_title=None,
@@ -152,8 +272,22 @@ class StoryRouteAgent:
         if decision.decision == "append_story":
             tid = decision.target_story_id
             if not tid or tid not in valid_story_ids:
+                fb = default_append_target_story_id(
+                    candidate_stories, story_meta, settings
+                )
+                if fb and fb in valid_story_ids:
+                    logger.info(
+                        "StoryRoute append 无效 target_story_id={}，回退默认 append {}",
+                        tid,
+                        fb,
+                    )
+                    return StoryRouteDecision(
+                        decision="append_story",
+                        target_story_id=fb,
+                        reason="invalid_target_default_append",
+                    )
                 logger.warning(
-                    "StoryRoute append 无效 target_story_id={}，回退 new_story",
+                    "StoryRoute append 无效 target_story_id={}，且无可用默认目标，回退 new_story",
                     tid,
                 )
                 return StoryRouteDecision(
@@ -204,4 +338,12 @@ class StoryRouteAgent:
         if not ok:
             logger.warning("StoryRouteAgent.plan_batch 校验失败: {}", err)
             return None
-        return plan
+        return normalize_batch_plan_reduce_new_story_fragmentation(
+            plan,
+            ordered,
+            chapter_category=chapter_category,
+            candidate_stories=candidate_stories,
+            valid_story_ids=valid_story_ids,
+            story_meta=story_meta,
+            settings=settings,
+        )

api/app/agents/memoir/story_route_payload.py

@@ -7,6 +7,7 @@ Story 路由：候选故事 JSON 载荷（summary 优先、预算裁剪、固定
 from __future__ import annotations
 
 import json
+import re
 from datetime import timezone
 from typing import Any, TYPE_CHECKING
 
@@ -15,6 +16,23 @@ if TYPE_CHECKING:
 
 from app.features.story.models import Story
 
+_PLAIN_SNIPPET_NOISE = re.compile(r"[`*_#]+")
+
+
+def _plain_opening_snippet_from_markdown(md: str, *, max_chars: int) -> str:
+    """无 summary 时供路由辨题的短文摘（弱化 Markdown 噪声）。"""
+    t = (md or "").strip()
+    if not t:
+        return ""
+    t = re.sub(r"!\[[^\]]*\]\([^)]+\)", "", t)
+    t = re.sub(r"\[([^\]]+)\]\([^)]+\)", r"\1", t)
+    t = re.sub(r"asset://\S+", "", t)
+    t = _PLAIN_SNIPPET_NOISE.sub("", t)
+    t = re.sub(r"\s+", " ", t).strip()
+    if len(t) <= max_chars:
+        return t
+    return t[: max_chars - 1] + "…"
+
 
 def _linked_chapters(s: Story) -> list[str]:
     links: list[str] = []
@@ -126,6 +144,9 @@ def _build_full_row(
     )
     if body:
         row["body_for_route"] = body
+    osnip = _plain_opening_snippet_from_markdown(canon, max_chars=260)
+    if osnip and len(osnip) >= 40:
+        row["opening_snippet"] = osnip
     return row