feat(comparison): 比价记录累计 LLM token(入/出)并在 admin 展示

server 从 llm_calls[].usage 累加 prompt/completion_tokens 落库,admin 比价记录列表/详情据此展示 token 数与估算成本;配套迁移 comparison_token_cols。

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

alembic/versions/comparison_token_cols.py

@@ -0,0 +1,37 @@
+"""add input_tokens / output_tokens to comparison_record
+
+比价记录增加本次 LLM 累计 token(server 从 llm_calls[].usage 累加),
+供 admin 比价记录页展示 token 数与估算成本。
+
+Revision ID: comparison_token_cols
+Revises: feedback_review_fields
+Create Date: 2026-06-23 00:40:00.000000
+
+"""
+from typing import Sequence, Union
+
+from alembic import op
+import sqlalchemy as sa
+
+
+# revision identifiers, used by Alembic.
+revision: str = "comparison_token_cols"
+down_revision: Union[str, Sequence[str], None] = "feedback_review_fields"
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    op.add_column(
+        "comparison_record",
+        sa.Column("input_tokens", sa.Integer(), nullable=True),
+    )
+    op.add_column(
+        "comparison_record",
+        sa.Column("output_tokens", sa.Integer(), nullable=True),
+    )
+
+
+def downgrade() -> None:
+    op.drop_column("comparison_record", "output_tokens")
+    op.drop_column("comparison_record", "input_tokens")

app/admin/schemas/comparison.py

@@ -33,6 +33,8 @@ class AdminComparisonListItem(BaseModel):
     step_count: int | None = None
     llm_call_count: int | None = None
     retry_count: int | None = None
+    input_tokens: int | None = None      # Σ usage.prompt_tokens(server 派生)
+    output_tokens: int | None = None     # Σ usage.completion_tokens(server 派生)
     device_model: str | None = None
     rom_vendor: str | None = None
     rom_name: str | None = None

app/api/v1/compare_record.py

@@ -77,8 +77,15 @@ def _backfill_llm_calls(record_id: int, trace_id: str) -> None:
         rec.llm_calls = calls
         rec.llm_call_count = len(calls)
         rec.retry_count = sum(1 for c in calls if c.get("error"))
+        # token 累加(usage 已被 pricebot llm_client 归一为 prompt/completion_tokens;
+        # error 的调用 usage 可能为 None,or {} 兜底)
+        rec.input_tokens = sum((c.get("usage") or {}).get("prompt_tokens") or 0 for c in calls)
+        rec.output_tokens = sum((c.get("usage") or {}).get("completion_tokens") or 0 for c in calls)
         db.commit()
-        logger.info("backfill llm_calls trace=%s n=%d", trace_id, len(calls))
+        logger.info(
+            "backfill llm_calls trace=%s n=%d in_tok=%d out_tok=%d",
+            trace_id, len(calls), rec.input_tokens, rec.output_tokens,
+        )
     except Exception as e:  # noqa: BLE001 best-effort
         logger.warning("backfill llm_calls failed trace=%s: %s", trace_id, e)
     finally:

app/models/comparison.py

@@ -124,6 +124,9 @@ class ComparisonRecord(Base):
     step_count: Mapped[int | None] = mapped_column(Integer, nullable=True)     # 客户端 agent loop 总步数
     llm_call_count: Mapped[int | None] = mapped_column(Integer, nullable=True) # 本次 LLM 调用次数(server 从 llm_calls 算)
     retry_count: Mapped[int | None] = mapped_column(Integer, nullable=True)    # LLM 失败重试次数(server 从 llm_calls error 算)
+    # 本次 LLM 累计 token(server 从 llm_calls[].usage 累加;旧记录/未采集为 None)
+    input_tokens: Mapped[int | None] = mapped_column(Integer, nullable=True)   # Σ usage.prompt_tokens
+    output_tokens: Mapped[int | None] = mapped_column(Integer, nullable=True)  # Σ usage.completion_tokens
     # 每次 LLM 调用明细 [{scene,model,input_messages,output,usage,latency_ms,error}];
     # server 收上报后按 trace_id 同机拉 pricebot 落库(见 compare_record 端点)。旧记录/未采集为 None。
     llm_calls: Mapped[list | None] = mapped_column(_JSON, nullable=True)