Фиксируем состояние

tests/pipeline_setup_v3/core/artifacts.py

@@ -66,8 +66,8 @@ class SummaryComposer:
             "",
             f"Passed: {passed}/{len(results)}",
             "",
-            "| File | Case | Mode | Query | Actual sub-intent | RAG layers | Pass |",
-            "|------|------|------|-------|-------------------|------------|------|",
+            "| File | Case | Mode | Query | Actual sub-intent | RAG layers | Tokens | Pass |",
+            "|------|------|------|-------|-------------------|------------|--------|------|",
         ]
         lines.extend(self._rows(results))
         failures = [item for item in results if not item.passed]
@@ -84,7 +84,7 @@ class SummaryComposer:
             rows.append(
                 f"| {item.case.source_file.name} | {item.case.case_id} | {item.case.mode} | "
                 f"{self._cell(item.case.query)} | {item.actual.get('sub_intent') or '—'} | "
-                f"{self._layer_text(item.details)} | {'✓' if item.passed else '✗'} |"
+                f"{self._layer_text(item.details)} | {self._token_text(item.details)} | {'✓' if item.passed else '✗'} |"
             )
         return rows
 
@@ -104,6 +104,16 @@ class SummaryComposer:
             return compact
         return compact[: limit - 1].rstrip() + "…"
 
+    def _token_text(self, details: dict) -> str:
+        token_usage = dict(details.get("token_usage") or {})
+        direct = token_usage.get("tokens_in_estimate")
+        if direct is not None:
+            return str(direct)
+        prompt = dict(details.get("diagnostics", {}).get("prompt") or {})
+        stats = dict(prompt.get("prompt_stats") or {})
+        value = stats.get("tokens_in_estimate")
+        return str(value) if value is not None else "—"
+
     def _llm_section(self, results: list[V3CaseResult]) -> list[str]:
         llm_results = [item for item in results if str(item.actual.get("llm_answer") or "").strip()]
         if not llm_results:

tests/pipeline_setup_v3/core/test_validators.py

@@ -2,6 +2,7 @@ from __future__ import annotations
 
 from pathlib import Path
 
+from tests.pipeline_setup_v3.core.artifacts import SummaryComposer
 from tests.pipeline_setup_v3.core.models import CaseExpectations, LlmExpectation, RetrievalExpectation, RouterExpectation, V3Case
 from tests.pipeline_setup_v3.core.validators import CaseValidator
 
@@ -66,3 +67,29 @@ def test_llm_contains_and_excludes_checks() -> None:
     }
 
     assert CaseValidator().validate(case, actual, {}) == []
+
+
+def test_summary_includes_token_usage_column() -> None:
+    case = V3Case(
+        case_id="llm-quality",
+        runner="agent_runtime",
+        mode="full_chain",
+        query="Где health endpoint?",
+        source_file=Path("cases.yaml"),
+    )
+    result_text = SummaryComposer().compose([
+        type(
+            "Result",
+            (),
+            {
+                "case": case,
+                "actual": {"sub_intent": "FIND_ENTRYPOINTS"},
+                "details": {"token_usage": {"tokens_in_estimate": 321}, "rag_rows": []},
+                "passed": True,
+                "mismatches": [],
+            },
+        )()
+    ])
+
+    assert "Tokens" in result_text
+    assert "321" in result_text