docs: update launch capacity and postgres tuning

docs/refactor/performance-benchmark-runbook.md

@@ -238,7 +238,7 @@ npm run perf:api:local
 npm run launch:gate
 ```
 
-`launch:gate` 会强制检查一条真实数据读路径压测证据：
+`launch:gate` 会强制检查真实数据只读压测和混合读写压测两条证据。只读证据用于基础上线门禁：
 
 ```json
 {
@@ -285,6 +285,37 @@ npm run perf:summary -- --input docs/refactor/performance-reports/api-benchmark-
 
 使用 `--allow-writes` 时，摘要工具输出 `capacityObservation`，不输出 `launchGateCheck`；不要把写入场景误填进生产上线门禁的 `performance.api-real-data-read`。
 
-更高的 50/100 并发、写入混合场景和容量结论仍应作为人工容量报告归档；门禁只负责挡住明显不达标的基础读路径。
+混合读写证据用于确认“创建练习 session -> 拉题 -> 答题 -> 交卷 -> 报告”的真实写入链路在上线容量基线内：
+
+```json
+{
+  "id": "performance.api-real-data-mixed",
+  "status": "pass",
+  "command": "PERF_START_SERVER=false PERF_API_BASE=https://api.example.com PERF_DURATION_SECONDS=120 PERF_CONCURRENCY=50 PERF_RAMP_SECONDS=15 PERF_INCLUDE_WRITES=true PERF_PRACTICE_FLOW_RATIO=0.1 PERF_AUTH_MODE=app_session npm run perf:api:local > docs/refactor/launch-artifacts/api-real-data-mixed-benchmark.log",
+  "completedAt": "2026-06-30T10:49:00+08:00",
+  "artifact": "launch-artifacts/api-real-data-mixed-benchmark.log",
+  "summary": {
+    "errors": 0,
+    "errorRate": 0,
+    "p95Ms": 0,
+    "p99Ms": 0,
+    "concurrency": 50,
+    "durationSeconds": 120,
+    "includeWrites": true
+  }
+}
+```
+
+混合读写上线门槛：
+
+- `errors = 0`
+- `errorRate <= 0.001`
+- `p95Ms <= 500`
+- `p99Ms <= 1200`
+- `concurrency >= 50`
+- `durationSeconds >= 60`
+- `includeWrites = true`
+
+更高的 100/150 并发和更长时间压测仍应作为人工容量报告归档；门禁负责挡住基础读路径和刷题写入闭环明显不达标的情况。
 
 证据中只记录报告路径、并发矩阵、P95/P99、错误率和结论，不保存真实 token、支付密钥、用户隐私或完整响应。