学习路径Learning path · 第 5 阶段Stage 5
S5 · 导出与发布
S5 · Ship it
这一步决定什么：让人真的玩到。AI 在这一步几乎帮不上忙 —— 这很正常。
What this stage decides: getting it in front of players. AI is almost no help here — and that is normal.
为什么 AI 在这里作用最小
导出是确定性工程：固定的流程、固定的工具链、明确的成功条件。
| 环节 | AI 能做什么 |
|---|---|
| 配置导出参数 | ✅ 能给模板 |
| 解决导出报错 | ⚠ 能猜，但报错信息往往不完整 |
| 平台审核合规 | ❌ 需要自己查 |
| 商店页面文案 | ✅ 能写 |
| 实际测试每种设备 | ❌ 做不了 |
⚠ 这不是 AI 的缺陷。导出是工程问题，工程问题的解法是查文档，不是让模型猜。
Why AI helps least here
Exporting is deterministic engineering: a fixed process, a fixed toolchain, an unambiguous success condition.
| Stage | What AI can do |
|---|---|
| Configure export settings | ✅ It can give a template |
| Fix an export error | ⚠ It can guess, but error messages are often incomplete |
| Platform review compliance | ❌ You must check this yourself |
| Store page copy | ✅ It can write it |
| Actually testing on each device | ❌ It cannot |
⚠ This is not a flaw of AI. Engineering problems are solved by reading the docs, not by asking a model to guess.
各平台的实际差异
⚠ 以下是我们的判断，具体条款请以平台官方为准。
| 平台 | 门槛 | ⚠ 常见的坑 |
|---|---|---|
| 网页（itch.io 等） | 最低 | 浏览器差异（Safari 的音频、WebGL 版本） |
| Windows / macOS | 中 | 签名与公证（macOS Gatekeeper） |
| 移动端 | 高 | 商店审核周期长；隐私合规 |
| 主机（Steam 等） | 最高 | 开发者账号费、审核、年龄分级 |
How platforms actually differ
⚠ What follows is our judgement; check the platforms' own terms for specifics.
| Platform | Barrier | ⚠ Common traps |
|---|---|---|
| Web (itch.io and similar) | Lowest | Browser differences (Safari audio, WebGL versions) |
| Windows / macOS | Medium | Signing and notarisation (macOS Gatekeeper) |
| Mobile | High | Long store review cycles; privacy compliance |
| Consoles (Steam and similar) | Highest | Developer account fee, review, age rating |
发布前必须自己测的
AI 帮不了、也必须做的：
- 在目标设备上真的跑过 —— 不是在开发机上
- 首次加载时间 —— 网页游戏尤其重要
- 音量/静音状态 —— ⚠ 忘记检查这个的开发者很多
- 存档是否真的保存 —— 清理浏览器数据后重试
- ⚠ 离线/断网行为 —— 是崩溃还是优雅降级？
What AI cannot do and you must
- You ran it on the target device — not just the dev machine
- First-load time — especially for web games
- Volume / mute state — ⚠ many developers forget this one
- Whether saves actually persist — retry after clearing browser data
- ⚠ Offline / no-network behaviour — does it crash or degrade gracefully?
⚠ 一个真实建议
先发一个「极简版」。
在 itch.io 上传一个能玩的、但素材很粗糙的版本，然后看有没有人玩。
⚠ 理由：这个行业的最大风险不是「游戏不好玩」，而是你不知道它好不好玩。
与其做完 80% 再发现方向错了，不如花 20% 做一个能验证方向的版本。
AI 让这个策略变得可行 —— 因为原型成本降到了可以「多发几个试」的程度。
⚠ One real piece of advice
Ship an extremely rough version first.
Upload something playable on itch.io, with deliberately crude assets, and see whether anyone plays it.
⚠ The reason: the biggest risk in this industry is not "the game is bad" — it is that you do not know whether the game is bad. Rather than discovering the wrong direction after 80% is done, spend 20% on something that validates it.
AI makes this strategy affordable — because prototype cost has dropped enough to "ship a few and try".
常见失败
- 追求「做完再发」 —— ⚠ 最常见的失败模式
- 发给朋友 —— ⚠ 朋友会给你礼貌性反馈，陌生人才给你真实反馈
- 只在开发机上测 —— 开发机的显卡和真实用户差得远
- 忽略 Safari / 老浏览器 —— 网页游戏的最大流失点
- 第一次发布就上商店 —— 审核周期会让你失去调整的窗口
Common failures
- Aiming to finish before shipping — ⚠ the most common failure mode
- Sending it to friends — ⚠ friends give you polite feedback; strangers give you real feedback
- Testing only on the dev machine — its GPU is nothing like a real user's
- Ignoring Safari / older browsers — the biggest leak for web games
- Going straight to a store on the first release — the review cycle removes your window to adjust
判断你这一步做完了吗
- 在非开发机上跑过
- 有人（不是你）玩过并给你反馈
- ⚠ 存档、离线、音量都测过
Are you done with this stage?
- It runs on a machine that is not your dev machine
- Someone who is not you has played it and given you feedback
- ⚠ Saves, offline behaviour and volume have all been tested