WandEnhancer 性能与成本：测量补丁、渲染器与面板开销

Define the measured surfaces

固定 Wand 客户端、fork revision、补丁选项、脚本清单、网络和 trainer 夹具，分别记录冷/热补丁、启动、脚本执行、3223 面板往返、CPU、内存、进程、端口和恢复时间，用 p50/p95 与人工维护成本衡量优化。

Do not mix first-run dependency downloads or antivirus scans with steady-state client behavior. Record cold and warm runs separately and keep the original client available so performance tests never become an irreversible upgrade experiment.

Capture latency and resource evidence

Measure patch duration, files changed, startup time, renderer script execution, panel connection, WebSocket round trip, and trainer action completion. Capture CPU, memory, process count, panel port state, and artifact size. Use p50/p95 over repeated disposable runs; one fast launch does not establish compatibility.

For custom scripts, measure each script independently and test the documented possibility that a script runs more than once per launch. A global guard can reduce duplicate work, but it should not hide a lifecycle bug or an expensive observer that remains attached after navigation.

Budget operational cost

The direct software cost is small, but operator time matters: fork maintenance, workflow minutes, malware scanning, client-version retesting, firewall support, and rollback. Keep a dated ledger of build duration, storage, support incidents, and time spent reviewing script changes rather than inventing a generic performance number.

Panel latency should be evaluated with the same LAN/VPN topology users will operate. A VPN may add round-trip time while reducing exposure; that is a security trade-off, not a reason to expose port 3223 publicly.

Optimize without weakening trust

Start with fewer renderer scripts, bounded observers, lazy panel assets, and explicit cleanup on disconnect. Re-run patch/restore, client startup, panel control, and antivirus review after each change. Accept an optimization only when the active trainer remains correct and every modified surface can still be removed.