fix: third-audit gate findings (explicit split, shadow explicit failures, N2-N5, N1 note)

Explicit gateway capability now splits known/unknown: known-missing
explicit models fail-fast under active (same shape as the gateway
default gate), unknown explicit models stay audit-only, shadow never
blocks; explicit priority now protects which model the user chose,
not that the chosen model must run. Shadow report reads
worker_capability_gateway_explicit rows as explicit_capability_failures
with a field note distinguishing the three explicit counts.
N2: coordinator requeue CAS rowcount=0 branch covered via a proxy
connection test (no refactor needed). N3: shadow endpoint test on
the create_server + real HTTP skeleton (capability, truncated, and
explicit-failure fields). N4: creation-layer rejection tests for
tools/instructions/tool_approvals on goal/coordinator tasks.
N5: model writeback failure logs a warning (best-effort kept).
N1: writeback capability-recheck trade-off documented in comment
only; recheck itself is a later batch.

cloud/server.py

@@ -4195,6 +4195,20 @@ class CloudControlPlaneHandler(BaseHTTPRequestHandler):
             ).fetchall()
             truncated = len(audit_rows) > 10000
             audit_rows = audit_rows[:10000]
+            # Audit fix 4（第三遍审计显式闭环）：显式能力检查
+            # 结果只记 worker_capability_gateway_explicit，影子
+            # 端点此前只读 worker_routing，运营方根本看不到它。
+            # 在此之外加读显式行，附为独立字段，不与已有两个
+            # explicit_n 口径混在一起（见报告里的字段说明）。
+            explicit_rows = conn.execute(
+                "SELECT detail FROM audit_log"
+                " WHERE project_id = ?"
+                " AND action = 'worker_capability_gateway_explicit'"
+                " ORDER BY created_at, id LIMIT 10001",
+                (project_id,),
+            ).fetchall()
+            explicit_truncated = len(explicit_rows) > 10000
+            explicit_rows = explicit_rows[:10000]
             task_rows = conn.execute(
                 "SELECT id, status, model FROM tasks"
                 " WHERE project_id = ?",
@@ -4229,6 +4243,51 @@ class CloudControlPlaneHandler(BaseHTTPRequestHandler):
             tasks_by_id, costs_by_task, explicit_n=explicit_n)
         report["project_id"] = project_id
         report["truncated"] = truncated
+        # 显式失败清单：解析 worker_capability_gateway_explicit
+        # 行（格式：task_id model=.. tier=.. mode=.. missing=..
+        # unknown=.. blocked=..；fix4 之前的行无 blocked 项，
+        # 按 False 计）。这与报告里两个 explicit_n 不同义：
+        # 顶层 explicit_n 是任务表里带模型字段的任务数；
+        # per-tier explicit_n 是 worker_routing source=explicit
+        # 审计行数；本字段只计网关显式能力检查的结果行，
+        # 含被拦的已知不满足与只审计的 unknown，blocked
+        # 字段区分两者。解析失败的行直接跳过、不谎报。
+        _explicit_failures = []
+        for _row in explicit_rows:
+            _detail = _row["detail"] or ""
+            _tokens = _detail.split()
+            if not _tokens:
+                continue
+            _kv = {}
+            for _tok in _tokens[1:]:
+                if "=" in _tok:
+                    _k, _v = _tok.split("=", 1)
+                    _kv[_k] = _v
+            if "model" not in _kv or "tier" not in _kv:
+                continue
+            _missing = _kv.get("missing") or "-"
+            _explicit_failures.append({
+                "task_id": _tokens[0],
+                "model": _kv["model"],
+                "tier": _kv["tier"],
+                "mode": _kv.get("mode") or "-",
+                "missing": [] if _missing == "-" else _missing.split(","),
+                "unknown": _kv.get("unknown") == "True",
+                "blocked": _kv.get("blocked") == "True",
+            })
+        report["explicit_capability_failures"] = _explicit_failures
+        report["explicit_capability_failures_truncated"] = \
+            explicit_truncated
+        report["explicit_counts_note"] = (
+            "explicit_n (top level) counts tasks whose model field is"
+            " set in the tasks table; per-tier explicit_n counts"
+            " worker_routing rows with source=explicit; "
+            "explicit_capability_failures lists"
+            " worker_capability_gateway_explicit rows only, i.e."
+            " gateway capability check outcomes for explicit models,"
+            " with blocked=True for known-missing models that were"
+            " refused and blocked=False for unknown models that were"
+            " audited but not blocked.")
         # Audit fix 3（I2，端点侧）：只读加载当前策略，
         # 给每档附能力状态（档模型 vs requires）。加载
         # 失败/未配置时状态为 policy_unavailable，不拦、