feat(api)!: memory single chain — async MemoryService, strict eval closure

Route all memory ingest/retrieve/enrichment/compaction through async MemoryService.
Remove legacy sync memory implementations (ingest/retrieve/compaction); Celery and
memoir Phase2 call asyncio.run into MemoryService-backed helpers.

Memoir Phase1 batch ingest uses MemoryService.ingest_transcripts_batch; drop chapters.
evidence_bundle_json mirror (Alembic 0015). Evaluation uses snapshot/link-only bundles;
raise EvidenceClosureMissing instead of partial/fallback lineage tiers.

Split memoir state into NarrativeCoverageState and InterviewControlState; delete the
_interview_meta_store adapter layer. Remove rolling-query and recent-fact fallback
settings from config and evidence assembly.

Update judges, docs, tests, and PlaygroundPage alignment.

Made-with: Cursor

api/app/agents/chat/interview_agent.py

@@ -31,7 +31,11 @@ from app.agents.chat.reply_limits import (
 )
 from app.agents.chat.reply_planner import maybe_refine_turn_plan_with_llm
 from app.agents.chat.stage_detection import keyword_fallback_primary_stage
-from app.agents.state_schema import MemoirStateSchema
+from app.agents.state_schema import (
+    MemoirStateSchema,
+    interview_control_state,
+    narrative_coverage_state,
+)
 from app.core.agent_logging import (
     agent_span,
     log_agent_payload,
@@ -154,14 +158,14 @@ class InterviewAgent:
             text_for_model = self._resolve_text_for_model(
                 user_message, normalized_user_message
             )
-            empty_slots = memoir_state.prompt_empty_slots_for_current_stage()
-            filled_slots = {
-                key: value.snippet
-                for key, value in memoir_state.slots.get(
-                    memoir_state.current_stage, {}
-                ).items()
-                if value.snippet
-            }
+            narrative_state = narrative_coverage_state(memoir_state)
+            control_state = interview_control_state(memoir_state)
+            empty_slots = control_state.prompt_empty_slots_for_stage(
+                narrative_state, memoir_state.current_stage
+            )
+            filled_slots = narrative_state.filled_slots_for_stage(
+                memoir_state.current_stage
+            )
             if detected_user_stage is not None:
                 du = detected_user_stage
             else:
@@ -173,7 +177,7 @@ class InterviewAgent:
             )
             recent_questions = extract_recent_questions(hw.window)
             conversation_turn_total = hw.turn_total
-            all_stages_coverage = memoir_state.all_stages_coverage()
+            all_stages_coverage = narrative_state.all_stages_coverage()
             persona = normalize_interview_persona(settings.chat_interview_persona)
             max_segments = int(settings.chat_interview_max_segments)
             max_tokens = int(settings.chat_interview_max_tokens)
@@ -406,7 +410,11 @@ class InterviewAgent:
         if not self.llm:
             return ["你好呀～ 又见面了。今天想从人生里哪一小段回忆开始聊聊？"]
         try:
-            empty_slots = memoir_state.prompt_empty_slots_for_current_stage()
+            narrative_state = narrative_coverage_state(memoir_state)
+            control_state = interview_control_state(memoir_state)
+            empty_slots = control_state.prompt_empty_slots_for_stage(
+                narrative_state, memoir_state.current_stage
+            )
             empty_slots_readable = [SLOT_NAME_MAP.get(s, s) for s in empty_slots]
             persona = normalize_interview_persona(settings.chat_interview_persona)
             prompt = get_opening_prompt(

api/app/agents/chat/interview_state_hints.py

@@ -5,13 +5,18 @@ from __future__ import annotations
 import re
 from collections.abc import Iterable
 
-# 与 `apply_duplicate_question_guard` 中整段替换句一致；用于判定是否需触发二次生成。
-DUPLICATE_QUESTION_GUARD_FALLBACK_ZH = "这一段我记住了。"
-
-from langchain_core.messages import AIMessage, BaseMessage, HumanMessage
+from langchain_core.messages import AIMessage, BaseMessage
 
 from app.agents.stage_constants import STAGE_DISPLAY_ZH, STAGE_SLOT_KEYS
-from app.agents.state_schema import KnownFact, MemoirStateSchema, PersonaThread
+from app.agents.state_schema import (
+    KnownFact,
+    MemoirStateSchema,
+    PersonaThread,
+    narrative_coverage_state,
+)
+
+# 与 `apply_duplicate_question_guard` 中整段替换句一致；用于判定是否需触发二次生成。
+DUPLICATE_QUESTION_GUARD_FALLBACK_ZH = "这一段我记住了。"
 
 _QUESTION_SPLIT_RE = re.compile(r"[?？]+")
 _SENTENCE_SPLIT_RE = re.compile(r"(?<=[。！？!?])")
@@ -220,10 +225,11 @@ def build_runtime_interview_state(
         )
 
     persona_additions: list[PersonaThread] = []
+    narrative_state = narrative_coverage_state(state)
     haystack = " ".join(
         [msg]
         + [fact.value for fact in state.known_facts[-8:]]
-        + list(state.filled_slots_for_stage(active_stage).values())[:4]
+        + list(narrative_state.filled_slots_for_stage(active_stage).values())[:4]
     )
     for trait, markers in _TRAIT_HINTS:
         for marker in markers:

api/app/agents/chat/personas.py

@@ -48,13 +48,3 @@ def get_interview_persona_tone_hint(persona: str) -> str:
         "短句像微信，一次最多一个具体问题，不重复上文已清楚的事；底色仍要**随和、像聊天伙伴**，别像考官盘问。"
         "允许用一两句场景感的短描写承接对方画面，不要只用干巴巴的确认句；情绪重时承接要有半句并肩，勿只回嗯。"
     )
-
-
-def get_interview_persona_block(persona: str) -> str:
-    """兼容旧名：返回空串，请改用 get_interview_persona_tone_hint。"""
-    return ""
-
-
-def get_opening_persona_line(persona: str) -> str:
-    """兼容旧名：与访谈轮次共用一句性格提示。"""
-    return get_interview_persona_tone_hint(persona)

api/app/agents/chat/stage_detection.py

@@ -4,7 +4,7 @@
 
 from __future__ import annotations
 
-from typing import Any, Optional
+from typing import Any
 
 from app.agents.chat.schemas import StageDetectionOutput
 from app.agents.chat.stage_prompts import (
@@ -23,12 +23,6 @@ from app.core.logging import get_logger
 
 logger = get_logger(__name__)
 
-
-def normalize_life_stage(raw: Optional[str], fallback: str) -> str:
-    """兼容旧名：统一走 normalize_chat_stage。"""
-    return normalize_chat_stage(raw, fallback)
-
-
 def keyword_fallback_primary_stage(user_message: str) -> str:
     """多阶段打分，取最高分；平局按 CHAT_STAGES 逆序优先（与历史 tie_order 派生一致，可能有小幅行为差异）。"""
     if not (user_message or "").strip():
@@ -100,5 +94,4 @@ __all__ = [
     "detect_primary_life_stage",
     "keyword_fallback_primary_stage",
     "life_stage_display_name",
-    "normalize_life_stage",
 ]