Week 4.5 Instructor Cheat Sheet

A two-page printable reference card for the Course 4.5 Advanced Workshop (Locked Tenant Reality) deck. Designed to sit next to the deck during the four-hour Teams session so you never have to scrub back through 40+ slides for a build phase spec, a tool-switch cue, or the next failure target. The reality-track companion to the Week 4 cheat sheet: same six modules, static-HTML execution layer, two CUI-authorized enterprise AI tools (genai.mil and Ask Sage) running side by side.

Print: Ctrl+P → Layout Portrait, Margins Default, Background graphics on. Two US Letter pages. Site chrome and this preview note are hidden in print.

Use: page 1 is the live-delivery card. Pacing strip, the two-tool frontier table to drop in chat, the Readiness Dashboard four-phase prompts, the two deliberate failure targets (O(n²) join in Phase 2, Chart.js CDN in Phase 3), and the Debug Clinic five-scenario rotation. Page 2 is the anchors you repeat all afternoon, recovery cues for when Ask Sage is unavailable or genai.mil misbehaves, the two-tool cross-check protocol, and the homework that gates Week 6.

4.5Course 4.5 · Advanced Workshop
(Locked Tenant Reality)
Instructor cheat sheet

Advanced Workshop (Locked Tenant Reality)

Frontier mapping across two tools, complex multi-component static-HTML build (Readiness Dashboard), Debug Clinic with five scenarios, two-tool QA cross-check, paired teach-back, workflow playbook with Which Tool column. Coach the re-prompt and the tool selection. Never type the fix.

4 hr · 6 modules
40+ slides · 1 break · 2 deliberate failure targets
Companion to week-4-5-advanced-reality.html + week-4-5-handouts.html
0:00M1 Frontier (two-tool)
0:30M2 Readiness Build
1:30Break 10′
1:40M3 Debug Clinic
2:20M4 QA + Cross-Check
2:50M5 Teach-Back
3:20M6 Playbook
130 min
Frontier Mapping (Two-Tool) 0:00,0:30 Silent + share · meta-frontier introduced
  1. Drop in chat (0:00): the two-column Tool Handles table (genai.mil vs Ask Sage) from the Week 4.5 hand-out pack. Set silent timer 15 min.
  2. Brief (read verbatim): "Two columns: what genai.mil chat handles in your domain, what Ask Sage with a reasoning model handles in your domain. Fifteen minutes silent, then we share, one row per builder. Picking the wrong tool is its own failure mode. We are naming it today."
  3. Introduce four classification categories: Frontier (AI capability limit), Platform (browser, OS, hosting), Context (prompt was insufficient), Meta-frontier (wrong tool for the job). The fourth is new for advanced builders.
  4. Share (15 min): one row per builder. When a Marine cites a recent failure, ask "frontier or meta-frontier?" before they answer with "the AI was bad."
Outcome · map Unit Advanced Frontier Map seeded One row per builder logged. Each map distinguishes genai.mil from Ask Sage. Restriction list captured separately for the program lead and the Capability Gap Map.
260 min
Complex Build, Readiness Dashboard 0:30,1:30 Four phases · mode + tool switching
  1. Hand-off (0:30): Deck → editor + Ask Sage + genai.mil in adjacent tabs. Pin the Readiness Dashboard reference build (builds/readiness-dashboard.html) in a side tab.
  2. Phase 1 (0:30,0:45) Centaur · Ask Sage with reasoning model: data architecture. Upload three sample CSVs (personnel, training, equipment). Ask for a join model, three edge cases, and the in-memory data shape. Constraint stated up front: single file, inline CSS+JS, no external scripts, no CDN.
  3. Checkpoint: every Marine names three edge cases out loud before they prompt again. Most rooms hit orphaned EDIPIs, NULL training, duplicate serials or expired certs.
  4. Phase 2 (0:45,1:00) Cyborg · student picks the tool: data ingestion + join. Student chooses genai.mil chat (fast iteration) or Ask Sage with the CSVs already uploaded. This is the first explicit tool-switch beat. Both paths are valid. Make them articulate the choice.
  5. DELIBERATE FAILURE TARGET (Phase 2, genai.mil path): the prompt asks for a join over 500+ records. Fast chat returns an array.find() O(n²) join that stalls, OR a JSON.parse without try/catch. Do not warn students. When hands go up, debrief on "frontier or context?" Answer: context. Fix: tell the AI it is getting 500+ records, ask for a Map keyed on EDIPI, wrap parse in try/catch.
  6. Phase 3 (1:00,1:20) Centaur: visualization. Readiness card, SVG bar chart by company, Not-Ready table.
  7. DELIBERATE FAILURE TARGET (Phase 3): fast chat returns a <script src> Chart.js CDN. Let the student catch the violation. If they do not, ask "Where is that chart actually coming from? Open DevTools, check Network." Re-prompt for inline SVG, vanilla JS, scarlet/gold/ink palette.
  8. Reasoning-vs-chat-model beat: name it out loud. The same Phase 3 prompt run through Ask Sage with a reasoning model honors the no-CDN constraint from the first response. Fast chat pushes verification onto the human. Neither is wrong. Pick deliberately.
  9. Phase 4 (1:20,1:30): verification + brief. Upload the generated readiness_output.csv and the original personnel_source.csv to Ask Sage. Ask it to find discrepancies. Then back to genai.mil to add the Commander's Snapshot button (read-only textarea + Copy).
  10. Coach in chat. If anyone skips the Ask Sage cross-check in Phase 4, stop them. "The cross-check IS the verification. Spot-check by hand only if the tool is down."
Deliverable · complex build Readiness Dashboard, single file, deployed Three sources joined, SVG chart rendered without a CDN, Not-Ready table accurate against source, Commander's Snapshot generates. Every builder articulates at least one mode switch AND at least one tool switch with a reason.
BRK10 min
Break. Be back at H+1:40. Pre-flight: pick the three Debug Clinic volunteers from chat (or queue scenarios from resources/debugging-scenarios.html); have the first volunteer's screen-share authorized; reset your genai.mil tab to a fresh conversation. 1:30,1:40
340 min
Debug Clinic (Five Scenarios) 1:40,2:20 Coach · fishbowl · 7 min per problem
  1. Hand-off (1:40): switch from your screen to a student's screen-share. You coach, you do not type.
  2. Protocol per problem (7 min total): Student presents in 2 min (expected, actual, what tried). Group diagnoses in 3 min. Instructor synthesizes the root-cause category in 2 min. Aim for 5 problems in 35 min.
  3. Five categories to name out loud: Frontier limitation, insufficient context, incorrect assumption, integration failure, data quality, OR tool-selection failure (meta-frontier).
  4. Five pre-built scenarios (fall-back if no volunteers): (1) Fetch race / double-submit, (2) Stale localStorage after schema change, (3) CSS specificity + timezone bug, (4) Event delegation on dynamic content, (5) Tool-selection failure (47 turns of context drift in genai.mil for a multi-file refactor that should have gone to Ask Sage from the start).
  5. Refuse the keyboard. "If I type it, you do not learn it. Re-prompt with these three changes." Walk away.
  6. Refuse the wrong tool. When the diagnosis is meta-frontier, the fix is to switch tools, not to re-prompt the same one harder.
Outcome · coaching habit Five live recoveries on the wall Each builder demonstrates re-prompt OR tool-switch, not rescue. At least one Scenario 5 (meta-frontier) sticks. The room sees coaching modeled five times.
430 min
QA Checklist + Two-Tool Cross-Check 2:20,2:50 Pairs · 7-min rotations · new protocol
  1. Drop in chat (2:20): the planted-error SOP excerpt from the hand-out pack (five errors: two fabricated references, one contradictory-timeline, one steps-out-of-order, one numbering-inconsistency). Pairs work in breakouts.
  2. QA Checklist (read once): Source verification (AI fabricates citations) · Data accuracy · Logic check · Format compliance · Domain review (smell test).
  3. NEW: Two-Tool Cross-Check Protocol. For high-stakes outputs (anything going to a CO, anything that becomes a system of record, anything that ends up in a FITREP), run the SAME prompt through both genai.mil AND Ask Sage with a reasoning model. Agreement raises confidence. Disagreement is a flag, not noise. Document systematic disagreements as frontier-map rows.
  4. Cost callout: the cross-check is not free. Two prompts + a comparison turn. Reserve it for outputs that will be acted on.
  5. Rotate every 7 min; reconvene at 2:47 for room scoring.
Outcome · verification 5 errors caught + cross-check protocol absorbed Pairs find all 5 in under 10 min. Each builder names one of their own past artifacts that should have gotten the two-tool cross-check.
530 min
Teach-Back 2:50,3:20 Pairs · 3-min teach + retell
  1. Reveal pairs drafted before the session (mix experience levels).
  2. Concept list (expanded for 4.5): Centaur vs Cyborg modes, Frontier mapping, Context-building, Iterative refinement, Verification protocols, The Jagged Frontier, Layer separation (Spec / Prototype / Production), Tool selection (genai.mil vs Ask Sage). The last two are new for 4.5.
  3. Brief (read verbatim): "Three minutes to teach the other person one concept from the curriculum. They retell it back. If they cannot, you have not taught it yet. The reality-track concept list adds Layer Separation and Tool Selection. Pick deliberately."
  4. Model the template once before students prepare. Concept + one-sentence definition + why it matters + real example from your work + common mistake + key takeaway.
  5. Force a deliverable if pairs go silent: "In ten minutes I want one slide each: the technique, when to use it, when not to."
Outcome · transferred skill Each builder taught one technique Required deliverable: one technique re-told back. Tool Selection and Layer Separation each chosen by at least one pair in the room.
630 min
Workflow Playbook (with Which Tool column) 3:20,3:50 Hand-off doc · tool selection is a design field
  1. Drop in chat (3:20): the Workflow Playbook blank template from the hand-out pack. NEW column: Which Tool (genai.mil / Ask Sage with reasoning model / both for two-tool cross-check).
  2. Show both worked examples first:
    • Example 1: Weekly Training Schedule on genai.mil chat, Cyborg mode. Fast iteration on a single-prompt task. ~45 min vs ~3 hr without AI.
    • Example 2: Weekly Readiness Rollup on Ask Sage with Claude Opus 4.7, Centaur mode. Three CSVs uploaded once, reasoning model holds schema across turns. ~1 hr vs ~4 hr without AI.
  3. Brief (read verbatim): "Pick one of your own recurring tools. Fill in: Task, Frequency, Mode, Which Tool, 4-8 Steps with Human/AI labels, Verification Checklist, Known Frontier Issues, Time Savings, Junior Development. The Which Tool field is not optional. Tool selection IS the workflow design."
  4. Capability Gap Map integration: each playbook generates at least two Gap Map rows tied to Production-Layer wishes (e.g., MOL write-back, automated brief delivery, direct connector to training currency).
  5. Issue the homework live (final 5 min): submit the playbook + 2 Gap Map rows to the unit shared library by EOW.
Deliverable · institutional One playbook + 2 Gap Map rows per builder First entry in the unit's reality-track playbook library. Which Tool field filled with rationale. Capstone-gate proposal due in 7 days.
EDD · Week 4.5 cheat sheet · page 1 of 2 · pacing, phase prompts, deliverables Print landscape off · portrait, default margins, background graphics on
4.5Course 4.5 · Advanced Workshop
(Locked Tenant Reality)
Cheat sheet · page 2

Anchors, recovery, homework

The phrases you repeat all afternoon, the failure modes to watch for when Ask Sage is down or genai.mil drifts, the two-tool cross-check posture, and the homework that gates Week 6 attendance.

Use during: M2 tool-switch beats
Use during: M3 meta-frontier scenario
Companion to week-4-5-advanced-reality.html

Pre-flight + hand-off cuesDone 48 h before; verified at T-30 min · chat-paste blocks staged for M1, M4, M6

  • Gate-check the roster. Every attendee has a deployed-tool URL on file from Course 3.5 (or Course 3 + at least one deployed reality-track build). Pull anyone without one to a 1:1 catch-up before M2.
  • Ask Sage access check. Confirm tenant access for each Marine 48 h out. If anyone is blocked, queue the contingency plan (genai.mil paste-and-summarize for the Ask Sage phases) and pre-stage a Capability Gap Map row: "Ask Sage access blocked for this unit."
  • Tabs and portals. Deck (Week 4.5), this cheat sheet, the Week 4.5 hand-out pack (Tool Handles table, SOP planted-error excerpt, playbook template), genai.mil, Ask Sage (signed in, reasoning model selected), a text editor, GitHub.com, the Readiness Dashboard reference build.
  • Sample data for M2. Three CSVs pre-staged in a OneDrive folder linkable from chat: personnel_source.csv (500+ rows), training_currency.csv, equipment_status.csv. Include the planted edge cases (orphaned EDIPIs, NULL training rows, duplicate serial numbers).
  • Bug bank. Keep three real prompt-failures from Week-3 or Week-3.5 chat ready for M3 if no one volunteers, including at least one tool-selection failure (Scenario 5 candidate).
  • Teach-back pairs drafted from the roster (mix experience levels). Reveal at the start of M5.
  • 0:00. Drop the two-column Tool Handles table into chat. Silent timer 15 min.
  • 0:30. Deck → editor + Ask Sage + genai.mil. Pin the Readiness Dashboard reference build in a side tab.
  • 0:45 · Phase 2 hand-off. Audibly: "Now you pick the tool. State your choice in chat before you prompt."
  • 1:40 · Debug Clinic. Switch from your screen to a student's screen-share. Coach only.
  • 2:20 · QA + Cross-Check. Drop the planted-error SOP excerpt. Pairs in breakouts; rotate every 7 min.
  • 3:20 · Playbook. Drop the blank template. Show both worked examples (Training Schedule on genai.mil, Readiness Rollup on Ask Sage) before students start their own.

Anchor phrasesPlant in M1; cite by name when you need them

  • 1. "Coach the re-prompt AND the tool selection. Never type the fix." The Debug Clinic test. If you took the keyboard, or you picked the tool for them, you broke the rule.
  • 2. "Frontier, context, platform, or meta-frontier?" The four categories. The fourth is new for 4.5. Cite at every failure debrief.
  • 3. Map the world as it is. Restrictions go in Outside. Captured separately for the program lead. Tool gaps become Capability Gap rows.
  • 4. Decompose before you prompt. Phase 1 of M2 is on paper (and Ask Sage with a reasoning model). No architecture, no help.
  • 5. Pick the tool on purpose. Fast chat for single-prompt drafts. Reasoning model for architecture and multi-file. Both for high-stakes cross-check. Defaulting to whichever tab is open is not a choice.
  • 6. Agreement raises confidence. Disagreement is a flag. The two-tool cross-check. Reserve for outputs that will be acted on.
  • 7. The teach-back is the test. If your partner cannot repeat it, you have not taught it.
  • 8. The playbook ships, not the prototype. Trigger, inputs, steps, verification, fallback, owner, which tool. The institutional artifact has a tool-selection field.
  • 9. HTML skills do not die when you turn in your CAC. The static-stack thesis. Cite when anyone asks why we are not on Power Platform today.

Recovery + pacing cuesWhen the room slips or a tool goes down, this is the order of operations

  • Ask Sage is not available for this unit today. Pivot every Ask Sage phase to genai.mil with paste-and-summarize. Surface the cost (token limits, no persistent context) explicitly. Add the Gap row. The pedagogy still works because the gap is felt directly.
  • genai.mil session resets mid-build. Start a new conversation. Paste the current state of the file plus a one-line "I am at Phase X fixing Y." Specificity beats restart.
  • github.io blocked from the network. Try GitLab Pages or Cloudflare Pages from the Static Hosting Cheat Sheet. If all blocked, switch to file:// for the day and add a Gap row.
  • No one volunteers for the Debug Clinic. Cold-call the bug bank. Run Scenario 5 (meta-frontier, 47 turns of context drift) as a fishbowl. That one is the highest-yield teaching moment.
  • A builder demands you type their fix during M3. Refuse. "If I type it, you do not learn it. Re-prompt with these three changes" OR "Switch tools and restate the goal once."
  • Frontier map turns into venting about tool restrictions. Re-anchor: "Map the world as it is, not as you wish it. Restrictions go in Outside. Specifics. Beneficiary counts. Time savings."
  • Phase 2 deliberate failure does not fire. Genai.mil sometimes gets the join right on the first try. If so, prompt a follow-up: "Now scale this to 5,000 records and add a malformed row to the input." The failure usually surfaces there.
  • Whole room is behind at 1:20. Compress Phase 4 verification to Ask Sage cross-check only (skip the Commander's Snapshot button). Protect M3 Debug Clinic and M6 Playbook. The playbook is the artifact that survives the week.
  • Pacing valve. Cut M3 to four scenarios (drop #4 Event Delegation, keep Scenario 5). Cut M5 teach-back to 2 minutes per pair. Never cut M6.

Homework to issue before logoffWeek 6 (Capstone) attendance gate hangs on the proposal

  • Polish the Readiness Dashboard. Keep the URL live. If you did not finish, ship the smallest defensible version by EOW. Post the URL in the Week 4.5 chat thread.
  • Submit the workflow playbook from M6 to the unit shared library by EOW. One per builder, with the Which Tool field filled and at least two Capability Gap Map rows attached. No exceptions.
  • Run one build through the EDD SOP QA process using the two-tool cross-check. Document one substantive disagreement between genai.mil and Ask Sage if one surfaces.
  • Document three failure cases with specifics: what failed, how you caught it, how you fixed it, which tool you used and whether the right move was switching tools. Log on the shared Advanced Frontier Map.
  • Run a teach-back at your section within 14 days. Three minutes, one technique, witnessed by your supervisor.
  • Identify one capability surprise. Somewhere AI performed better than you expected, on either tool.
  • Capstone gate. To attend Week 6 (Full-Stack), submit a brief proposal naming the problem you will ship end-to-end and the tool selection rationale. Due 7 days before Week 6. No proposal, no seat.
EDD · Week 4.5 cheat sheet · page 2 of 2 · anchors, recovery, homework Pair with the Week 4.5 Facilitator Pack and the Week 4.5 hand-out pack