feat(eval): memoir A/B chapter judging and eval-web parity with dialogue

- Judge baseline excerpt and library chapter separately; build_memoir_compare_summary for gate, nine-dim and leaf deltas.

- Memoir SSE chapter payload: baseline_judge, compare_summary, baseline_judge_error.

- MemoirJudgeOutput: loose score coercion and post-validate clamp; memoir judge prompt caps from settings.

- app-eval-web: two-column MemoirScoreCard layout, MemoirCompareSummary, chapter blocks and CSS.

- Add memoir_compare_summary, log_events, celery_log_context, memoir_pipeline_progress; tests and migration 0014.

- Misc: memory/evidence and enrichment paths, task/orchestrator updates, internal-eval docs, env examples.

api/app/tasks/story_title_tasks.py

@@ -0,0 +1,134 @@
+"""Async story title refinement after new story create (placeholder first)."""
+
+import time
+
+from celery import shared_task
+
+from app.core.db import get_sync_db
+from app.core.dependencies import get_llm_provider
+from app.core.logging import get_logger
+
+logger = get_logger(__name__)
+
+
+@shared_task(bind=True, max_retries=2, default_retry_delay=15)
+def generate_story_title_after_create(
+    self,
+    story_id: str,
+    chapter_category: str,
+    oral_scope: str,
+    user_id: str,
+):
+    """Replace placeholder title with LLM title when body is long enough."""
+    from app.agents.chat.prompts_profile import format_user_profile_context
+    from app.agents.memoir.narrative_agent import NarrativeAgent
+    from app.features.memoir.state_service import get_or_create_state_sync
+    from app.features.memoir.story_pipeline_sync import (
+        _maybe_generate_title,
+        _placeholder_title,
+        _slot_snippets_for_narrative,
+    )
+    from app.features.story.models import Story
+    from app.features.user.models import User
+
+    t0 = time.perf_counter()
+    logger.info(
+        "event=story_title_task_start story_id={} user_id={} chapter_category={} "
+        "msg=故事标题精修任务开始",
+        story_id,
+        user_id,
+        chapter_category,
+    )
+    try:
+        with get_sync_db() as db:
+            st = db.get(Story, story_id)
+            if not st or str(st.user_id) != str(user_id):
+                ms = (time.perf_counter() - t0) * 1000
+                logger.info(
+                    "event=story_title_task_skip story_id={} reason=not_found duration_ms={:.1f} "
+                    "msg=标题精修跳过（故事不存在或无权限）",
+                    story_id,
+                    ms,
+                )
+                return {"status": "skip_not_found"}
+            expected_ph = _placeholder_title(chapter_category)
+            if (st.title or "").strip() and (st.title or "").strip() != expected_ph:
+                ms = (time.perf_counter() - t0) * 1000
+                logger.info(
+                    "event=story_title_task_skip story_id={} reason=user_modified duration_ms={:.1f} "
+                    "msg=标题精修跳过（用户已改标题）",
+                    story_id,
+                    ms,
+                )
+                return {"status": "skip_user_modified"}
+
+            llm = getattr(get_llm_provider(), "langchain_llm", None)
+            if not llm:
+                ms = (time.perf_counter() - t0) * 1000
+                logger.info(
+                    "event=story_title_task_skip story_id={} reason=no_llm duration_ms={:.1f} "
+                    "msg=标题精修跳过（无 LLM）",
+                    story_id,
+                    ms,
+                )
+                return {"status": "skip_no_llm"}
+
+            user_obj = db.get(User, user_id)
+            user_profile = ""
+            birth_year = None
+            if user_obj:
+                birth_year = user_obj.birth_year
+                user_profile = format_user_profile_context(
+                    birth_year=user_obj.birth_year,
+                    birth_place=user_obj.birth_place,
+                    grew_up_place=user_obj.grew_up_place,
+                    occupation=user_obj.occupation,
+                )
+
+            state = get_or_create_state_sync(user_id, db)
+            slot_snippets = _slot_snippets_for_narrative(
+                state=state,
+                chapter_category=chapter_category,
+                user_id=user_id,
+            )
+            md = (st.canonical_markdown or "").strip()
+            new_title = _maybe_generate_title(
+                NarrativeAgent(),
+                chapter_category=chapter_category,
+                md=md,
+                slot_snippets=slot_snippets,
+                user_profile=user_profile,
+                user_birth_year=birth_year,
+                llm=llm,
+                oral_scope=oral_scope or "",
+            )
+            if not new_title.strip() or new_title.strip() == expected_ph:
+                ms = (time.perf_counter() - t0) * 1000
+                logger.info(
+                    "event=story_title_task_skip story_id={} reason=placeholder duration_ms={:.1f} "
+                    "msg=标题精修跳过（仍为占位）",
+                    story_id,
+                    ms,
+                )
+                return {"status": "skip_placeholder"}
+            st.title = new_title
+            db.commit()
+            ms = (time.perf_counter() - t0) * 1000
+            logger.info(
+                "event=story_title_task_done story_id={} user_id={} duration_ms={:.1f} "
+                "msg=故事标题精修完成",
+                story_id,
+                user_id,
+                ms,
+            )
+            return {"status": "ok", "title": new_title}
+    except Exception as exc:
+        ms = (time.perf_counter() - t0) * 1000
+        logger.warning(
+            "event=generate_story_title_after_create_failed story_id={} duration_ms={:.1f} err={} "
+            "msg=故事标题精修失败",
+            story_id,
+            ms,
+            exc,
+        )
+        raise self.retry(exc=exc) from exc