Print this page in landscape from Chrome or Edge: Ctrl+P → Layout Landscape → Margins None or Default → Background graphics on. The site chrome above is hidden in print.

In the print dialog, leave Margins at Default (or pick None). Picking Custom or Minimum shrinks the printable area below what this layout assumes and the handout tips onto a second page.

4.5 Course 4.5 · Student Handout

Advanced Workshop (Locked Tenant Reality)

Take this home. Three Layers, Frontier classification, tool-selection cheat sheet, plus seed rows for the Capability Gap Map and your Workflow Playbook.

Duration4 hr (1 break)
AudienceExperienced builders
PrereqCourse 3.5 + 1 deployed tool
0:00,0:30M1 Frontier Mapping30 min · build map
0:30,1:30M2 Complex Build60 min · Readiness Dashboard
1:30,1:40Break10 min
1:40,2:20M3 Debugging Clinic40 min · pair
2:20,2:50M4 QA & Verification Drill30 min · timed
2:50,3:20M5 Teach-Back30 min · breakouts
3:20,3:50M6 Workflow Playbook30 min · build
3:50,4:00Wrap10 min

The Three LayersWhere we work, where we can't, and what to write down

The locked tenant doesn't change the work. It changes which layer you reach for first. Spec is where ideas get shaped, Prototype is where they get a URL, Production is where the wall is. Every Course 4.5 build moves through all three.

  • Spec layer. genai.mil for single-shot chat and short reasoning. Ask Sage for multi-file uploads, longer context, and dataset work. Both are first-class. Pick the one whose limits don't matter for your task.
  • Prototype layer. Single-file static HTML on a public host (GitHub Pages, Cloudflare Pages, Netlify). A URL another Marine can open on their phone. This is where Course 3.5 deliverables live.
  • Production layer. The wishlist. Power Platform inside the locked tenant, official systems of record, anything that needs CAC-gated auth. Write the gap down on the Capability Gap Map, then ship the prototype anyway.
  • Rule. Work the layers you can reach. Document the wall on the layer you can't. Don't fake a production answer with a prototype.
  • Mode-switching across layers. Start cyborg on Spec to discover the shape of the problem. Move to centaur on Prototype to verify before you sign anything. Production is a separate conversation, not the next step in the same prompt.
  • Hosting check. Open your deployed prototype from the duty workstation before class. If GitHub Pages, Cloudflare Pages, or Netlify is blocked, raise it in M1 so we can pick a fallback before the build clock starts.

Frontier classificationThree buckets to sort every task into

Every task in your section sits in one of three buckets. Knowing which bucket you're in tells you how hard to verify, which tool to reach for, and whether to bet a deliverable on the result. Sort before you prompt.

  • ContextInside the frontier. AI does this well today with normal prompting. Reach for the tool, verify the output, ship. No drama.
  • FrontierOn the edge. AI gets it right sometimes. You only catch the wrong answers if you already knew the answer. Use centaur mode and verify at every phase boundary.
  • Meta-frontierWrong tool reached for. The AI's actual capability is fine; you're using the wrong one for this kind of problem (genai.mil chat for a multi-file refactor that needs Ask Sage + reasoning, or vice versa). The fix is tool selection, not better prompting. See Module 3 Scenario 5.
  • Sorting ruleIf you can't tell which bucket a task is in, treat it as frontier and verify. Cost of over-verifying a context task is small. Cost of under-verifying a frontier task is a wrong answer that ships.
  • Date your mapThe frontier moves. A meta-frontier task today is a context task two quarters from now. Write the date next to every row. Re-sort quarterly.
  • Surprise both waysLog capability surprises (AI did better than you expected) in the same map as failure surprises. Both signals are how you keep the map honest.

Tool-selection cheat sheetWhich Spec-layer tool, when

Both genai.mil and Ask Sage are first-class Spec-layer tools. Different shapes. Pick by the task in front of you, not by habit. The four rules below cover 90 percent of decisions.

Single-shot chat, short reasoning, fast iteration → genai.mil. Drafting a paragraph, sanity-checking logic, sketching an outline, a regex or date math one-liner, a small targeted code change. CAC sign-in, no setup. The default for centaur whiteboard sessions and quick cyborg sketches. Multi-file refactors of a static-stack tool are painful here; that's an Ask Sage task.

Multi-file, dataset, long context, attached references → Ask Sage. A real folder of source docs. A CSV you want sliced. An SOP you want cross-checked against three policy PDFs. Larger context window, file handling, model choice. The default when "paste it into the prompt" is laughable.

High-stakes output (briefs, formal correspondence, anything signed) → both, deliberately. Draft in one, verify in the other. Different models catch different errors. The five QA protocols (source, logic, format, numbers, substance) apply either way. If the output is going to the CO, run both.

If your tool of choice is down → the other one is the fallback, not "wait until Monday." Both work inside the tenant. Both are sanctioned. Knowing both means you don't get stuck.

Switching costs you a minute, not a session. If you find yourself jamming a multi-file task into a single-shot prompt, that's the signal. Stop, open the other tab, paste the files. The five QA protocols catch the rest.

Anchor phraseThe tool isn't the skill. Knowing when to switch is.

Capability Gap Map seedTwo rows to fill before you leave

How to write a row. Name the specific production-layer capability that would 10x one of your tools. Name the specific permission, system, or policy that blocks it. Name the workaround you're using today. Specific is better than tidy.

  • Row 1, build: __________________ · capability needed: __________________ · blocker: __________________ · workaround: __________________
  • Row 2, build: __________________ · capability needed: __________________ · blocker: __________________ · workaround: __________________
  • What counts as a real blocker. A named policy, permission, or system, not "the tenant is locked." Examples: SharePoint app-only API access denied, Power Platform premium connector behind a paywall, no outbound HTTPS from the duty workstation.
  • What counts as a real workaround. A path you'd actually walk a junior Marine through. "Copy-paste from the CAC-gated portal once a week" counts. "I'll figure it out" doesn't.
  • Quarterly consolidation. Instructor rolls every student's rows into a single matrix and forwards to the HQMC AI working group. Your row is the unit's voice.

Workflow Playbook seedOne row, seven fields, one recurring task

The Workflow Playbook is the graduation deliverable. Not the tool, the playbook. A one-page sheet a relief in place can run without calling you.

Pick one real, recurring task from your job. Fill the seven fields below before you leave Module 6. Specific enough that a junior Marine could run it without you. That's the graduation deliverable.

  • Task: __________________
  • Frequency & mode: __________________ · (centaur / cyborg / mixed)
  • 4 to 8 H/AI steps: __________________
  • Verification checklist (3 to 5 items): __________________ · if your playbook has none, it isn't done.
  • Known frontier issues: __________________ · Time savings: __________________ · Junior development note: __________________
  • The 201 multiplier. You owe two Marines forward. Calendar the first teach-back before your next PCS, not after. The unit's capability compounds when the playbook runs without you.
  • Run it once before you teach it. End-to-end, with the real task, on a real workday. If a step breaks, fix it before the playbook leaves your hand.

URLs to come back toBookmark these on the workstation you build from

These are the post-session references. Open them from your duty workstation, not your phone. If a URL doesn't load from a duty laptop, that's a Capability Gap Map row right there.

  • Course 4.5 page: /courses/advanced-reality.html · agenda, prereqs, the deck, the live in-class handout.
  • Reference build (Readiness Dashboard): /builds/readiness-dashboard.html · the Module 2 build target. Open it from your duty workstation as a hosting check before class.
  • Tool-selection guide: /resources/tool-selection-guide.html · the long-form version of the cheat sheet above. genai.mil vs Ask Sage, by task type. Includes the Ask Sage model picker.
  • Marine Prompt Library: /resources/marine-prompt-library.html · 35 battle-tested prompts for awards, counseling, FITREP, MCO interpretation, OPSEC scrub, DD-form drafting. Paste and ship.
  • Tips and Tricks: /resources/tips-and-tricks.html · model-switching, context window discipline, file ingestion, the 4-line prompt template.
  • Citation Verification: /resources/citation-verification.html · the protocol that stops you from signing a document with a hallucinated MCO citation.
  • Debugging scenarios: /resources/debugging-scenarios.html · the five Module 3 scenarios with answer keys. Use them for informal team debriefs.
  • Capability Gap Map: /resources/capability-gap-map.html · the template. Add your rows here after class. Quarterly export to HQMC.
  • Static hosting cheat sheet: /resources/static-hosting-cheat-sheet.html · GitHub Pages, Cloudflare Pages, Netlify. The 10-minute path from .html on disk to a public URL.
  • SOP (PDF): /pdf/SOP_Expert_Driven_Development_v5.pdf · the verification checkpoints and the five QA protocols, in one document.
  • All handouts index: /handouts/ · every course's take-home in one place. Print the ones for the courses you're running next.

Power Apps skills die when you turn in your CAC. HTML skills don't. Knowing when to switch tools makes both skills last.