fix: gate INFRA isolation edge cases (all-INFRA fail-loud, parse_error in guard)

app/harness/validate.py

@@ -332,7 +332,9 @@ async def _resolve_baseline_block(
     if miss_units:
         miss_questions = flatten_units(miss_units)
         r_b = await run_inference(miss_questions, run_id=run_id, skills_dir=base_skills_dir)
-        errors_inc = r_b.stop_reason_counts.get("error", 0)
+        # 护栏错误计数与 INFRA 判定口径一致：error + parse_error 都计入，
+        # 使 parse_error 风暴同样能触发 gate_guard_err 熔断（不被绕过）。
+        errors_inc = sum(r_b.stop_reason_counts.get(reason, 0) for reason in _INFRA_STOP_REASONS)
         denom_inc = r_b.total
         infra_qids = _infra_question_ids_from_db(log, r_b.run_id, miss_questions)
         fresh_per_q = _candidate_correctness_from_db(log, r_b.run_id, miss_questions)
@@ -380,7 +382,9 @@ async def _run_candidate_block(
     questions = flatten_units(units)
     r_c = await run_inference(questions, run_id=run_id, skills_dir=cand_dir)
     c_per_q = _candidate_correctness_from_db(log, r_c.run_id, questions)
-    return c_per_q, r_c.stop_reason_counts.get("error", 0), r_c.total
+    # 护栏错误计数与 INFRA 判定口径一致：error + parse_error 都计入。
+    errors_inc = sum(r_c.stop_reason_counts.get(reason, 0) for reason in _INFRA_STOP_REASONS)
+    return c_per_q, errors_inc, r_c.total
 
 
 def _build_evidence_rows(
@@ -624,6 +628,15 @@ async def _run_local_validation(
             log=log,
             run_id=f"{gate_run_prefix}_b{block_idx}_base",
         )
+        # 本块全 INFRA：无有效单元可配对——候选无需空跑，仅把基线侧错误计入护栏后
+        # 累计剔除数进入下一块（护栏仍能在整轮 INFRA 错误率超阈值时熔断）。
+        n_excluded += len(unit_chunk) - len(valid_chunk)
+        if not valid_chunk:
+            errors += err_b
+            infra_denom += den_b
+            _check_infra_guard(errors, infra_denom, gate_guard_err)
+            continue
+
         # 候选侧只跑基线侧判定有效（非 INFRA）的单元，保证配对 unit_ids 两侧一致
         c_per_q, err_c, den_c = await _run_candidate_block(
             units=valid_chunk,
@@ -639,11 +652,6 @@ async def _run_local_validation(
         infra_denom += den_b + den_c
         _check_infra_guard(errors, infra_denom, gate_guard_err)
 
-        # 本块全 INFRA：无有效单元可配对，累计剔除数后跳过判定进入下一块
-        n_excluded += len(unit_chunk) - len(valid_chunk)
-        if not valid_chunk:
-            continue
-
         # Phase 3: 折叠成单元视图 + 配对 + 证据行 + 块间判定（均用有效单元）
         c_units = unit_correctness_view(valid_chunk, c_per_q)
         candidate_per_q.update(c_per_q)
@@ -668,8 +676,11 @@ async def _run_local_validation(
         if verdict.decision != "continue":
             break
 
+    # verdict 仍为 None ⟺ 全部单元被 INFRA 排除（空 ladder 已在入口拒绝）。
+    # 明确失败，避免落到误导性的"空阶梯"断言而无法定位为 INFRA 原因。
+    if verdict is None:
+        raise RuntimeError("gate 阶梯所有 unit 被判为 INFRA 排除，无法验证（检查推理基础设施）")
     # 最后一块判定即终态（n_remaining=0 → provisional/inertia）
-    assert verdict is not None, "空阶梯应已在 validate_skill_local 入口拒绝"
     return _finalize_outcome(
         verdict=verdict,
         w=w,