Two samples of the same subject at two scales — the clearest way to see what the scale actually changes. All case materials are fabricated; nothing here comes from a real client engagement.
同一个主题的两档样例，正好用来看分档到底差在哪。所有案例素材均为虚构， 不含任何真实客户材料。
| Example | Scale · 档位 | Contents · 内容 |
|---|---|---|
ai-agent-insurance |
S · Talk / 分享课 | 60 min, 29 slides. Outline + deck only. The course on the live demo. |
ai-agent-insurance-workshop |
M · Workshop / 内训课 | Full day, 31 slides, plus a design blueprint and the full delivery package. The pack on the live demo. |
deck.html is a build product and is not committed — one command regenerates it.
The delivery package is committed twice on purpose: package/*.md is the source you
edit, package/html/ is the rendered deliverable a client actually receives. The HTML is
in the repo so you can double-click and see what comes out — including how it paginates
when printed — without running anything. Never hand-edit the HTML: change the md and
re-run package --render.
deck.html 是构建产物，不入库，一条命令即可生成。
交付包刻意存两份：package/*.md 是你改的源，package/html/ 是客户实际拿到的交付物。
HTML 入库是为了让人 clone 下来双击就能看见成品长什么样、打印时怎么分页，不必先跑命令。
但永远不要手改 HTML——改 md，重新跑 package --render。
# S · 分享课
cp -R examples/ai-agent-insurance courses/
node nextcourse.js render ai-agent-insurance
# M · 内训课
cp -R examples/ai-agent-insurance-workshop courses/
node nextcourse.js check ai-agent-insurance-workshop # 0 errors, 0 warnings
node nextcourse.js render ai-agent-insurance-workshop # deck.html
node nextcourse.js package ai-agent-insurance-workshop --render # package/html/
node nextcourse.js export ai-agent-insurance-workshop --with-packageNot the deck. 不是那份课件。
| File | Why it matters |
|---|---|
course.meta.md |
The build contract: frontmatter (outcomes, scale, duration) + the page-level outline. build.js reads only this. |
course.blueprint.md (M) |
The design layer: needs diagnosis, what this run will not cover, objectives incl. Kirkpatrick, the day's timeline, the module list with duration / activity / deliverable / outcomes covered. |
package/7_alignment.md (M) |
Generated by check. Outcome × module × activity × deliverable × evidence — the whole backward-design loop in one table. |
package/3_rubric.md (M) |
Every cell is an observable behaviour. No "good", no "fair", no "needs improvement". |
package/html/0_index.html (M) |
What the client receives. The number prefixes are the delivery order — a file manager sorts the folder into the order a client should read it. Open it, then hit print preview. |
nextcourse check ai-agent-insurance-workshop passes with 0 errors, 0 warnings,
0 notes — every outcome is taught by a named module and proven by named evidence,
every module with an activity has an Activity page in the deck, module durations add up
against the declared day, and the blueprint carries all eight sections.
That is the bar the tool is built to enforce, not a bar it hopes you clear.
M 档样例跑 check 是 0 error / 0 warning / 0 note：每条成果都有模块教到、有证据测到，
每个有活动的模块都有对应的活动指令页，模块时长与声明总时长对得上，蓝图八节齐全。
这是工具会强制的下限，不是它希望你自觉达到的高线。
The workshop sample deliberately describes tool categories ("office agent", "agent framework") instead of naming products. Tools get replaced every six months; teaching the judgement outlives teaching the buttons — and a sample course shouldn't read as an endorsement.
M 档样例刻意只写工具品类（「办公智能体」「agent 框架」），不写具体产品名： 工具半年一换，教品类判断比教按钮位置耐用，样例课也不该读起来像在替谁背书。