Platform Training (Locked Tenant Reality)
Three deployed static-HTML tools in four hours. Centaur, cyborg, and your own problem. Plus a Capability Gap Map row for HQMC.
Bring with youDone before you walk in
- Course 2 prototype with notes on what worked, what didn't, and one failure case. Failure sharing is the first 10 minutes after Build #1.
- genai.mil access verified. Sign in on the workstation you will use. Test starting a conversation.
- Free GitHub account. Sign up at
github.comif you don't already have one. Personal email works. Two-factor takes 10 min the first time. - Text editor. Notepad, Notepad++, or VS Code. Anything that saves a plain
.htmlfile. - Hosting check. Open the reference build URL from your duty workstation before class. If it loads, you are good. If not, talk to the instructor about file:// fallback.
- Build #3 problem statement. One sentence on tool + user. You will declare it out loud at 2:40.
Key termsThe vocabulary you will hear today
- The Three LayersSpec (genai.mil), Prototype (static HTML on a public host), Production (the wishlist). We work the layers we can reach. We document the wall on the layer we can't.
- Phase boundary (centaur)A non-negotiable checkpoint between design, build, and verify. You cross only when the previous phase is complete and signed.
- Whiteboard firstEvery input, business rule, and edge case written down before AI runs a single prompt. Centaur opener.
- Deliberate failureA point in the build where AI is known to give wrong instructions. You let it happen so you learn to spot it.
- Frontier recognitionCatching a wrong AI answer fast enough to course-correct in the same prompt, not three prompts later.
- Capability Gap MapThe second artifact of the course. For each build, name the production-layer capability that would 10x the tool and the specific permission that blocks it. Forwarded to HQMC.
Exercises in classThree deployed builds, two artifacts
Build #1 · Centaur (65 min), Watch Bill Builder. Section leader's recurring Tuesday-afternoon problem. Rotation through the available roster, skip anyone on leave, fairness check, prints clean, copies clean into Outlook. Single HTML file. Phase 1 (20 min) is whiteboard only, no AI. Done: a 14-day bill prints clean, the email body pastes into Outlook intact, every edge case errors instead of silently doing nothing, URL openable on a phone.
Failure share (10 min, post-Build #1). Bring one failure case from Build #1 to the room. We capture patterns. Specific is better than tidy.
Build #2 · Cyborg (60 min), TEEP Builder. The S-3 brief slide is stale by Friday. Build a TEEP that lives in the browser, persists across sessions, color-codes status, and produces a current S-3 brief paragraph on demand. You will restructure something mid-build, that is the point. Done: a working TEEP with localStorage persistence, class-based status coloring you can explain, and a copy-pasteable brief.
Build #3 · Your problem (50 min), Declare first. One sentence: tool type + user. One mode + why. Two specific frontier risks. If you can't name two specific risks, revise before you start. Done at 3:25: simplest version of your tool runs, deployed to a public URL.
M6 · Frontier Map + Capability Gap Map (20 min). Add at least one row to each map. Frontier Map names a specific task type where AI surprised you. Capability Gap Map names the specific production-layer permission that would 10x one of your builds.
Anchor phrasePower Apps skills die when you turn in your CAC. HTML skills don't.What you'll be able to doBy the end of the session
- Build a single-file static-HTML tool in centaur mode with verification at every phase boundary.
- Build a single-file static-HTML tool in cyborg mode, including a deliberate restructure mid-build.
- Choose between modes for your own problem, defend the choice, and ship in 50 minutes.
- Deploy to a public URL another Marine can open on their phone.
- Recognize a wrong AI answer fast enough to course-correct in the same prompt.
- Write a Frontier Map row and a Capability Gap Map row that another Marine could act on.
HomeworkRequired to attend Week 4 Advanced Workshop
- Polish Build #3 to a deployable state. Keep the URL live.
- Run one build through the EDD SOP QA process. Don't skip the verification checkpoints, they are the entry ticket.
- Document three specific failure cases from finishing. Log them on the shared frontier-map doc.
- Add at least two Capability Gap Map rows. Real, specific, named. The instructor consolidates quarterly to HQMC AI working group.
- Identify one capability surprise. Somewhere AI performed better than you expected.
- If you get blocked, contact the program lead before Week 4. We'd rather unblock you than have you skip the workshop.