How to use this page
Read it once before you teach. During the live exercise, drive the room from the deck's debrief slides — not from this page — so the count happens before the answer goes up. The "Slide" column tells you which debrief card you'll reveal each error from.
Doc 1 errors — mapped to Slide 33
Open the document: Doc 1 · Award narrative. Open the debrief slide: Week 1 deck · Slide 33.
| Where in the doc | Planted error | Debrief card |
|---|---|---|
| Para 2 | "47% reduction in gear-deficiency findings" — confident percentage with no source. AI loves false precision; awards boards will ask. | Slide 33 · "What's wrong" · bullet 1 |
| Para 2 | "156 man-hours per quarter" — specific man-hours claim with no underlying calculation. Sounds verifiable; isn't. | Slide 33 · "What's wrong" · bullet 1 |
| Ref (a) | "MARADMIN 045/26" — cited MARADMIN that does not exist. Cited reference numbers in confident, plausible formats are a common AI fabrication. | Slide 33 · "What's wrong" · bullet 2 |
| Para 3 | "Highest mark in the regiment" — bold superlative, no source. Sign one and you've signed an unverifiable claim. | Slide 33 · "What's wrong" · bullet 3 |
| Paras 1, 2, 5 | Filler vocabulary — unparalleled, tireless (twice), selfless. Sound impressive, prove nothing. | Slide 33 · "What's wrong" · bullet 4 |
What's right (don't strike it): NAM narrative format, period of service and unit placement, "reflected great credit upon" closing boilerplate. That's the lesson under the lesson — AI gives you a real starting structure, then fills it with unsupported claims.
Doc 2 errors — mapped to Slide 34
Open the document: Doc 2 · SOP excerpt. Open the debrief slide: Week 1 deck · Slide 34.
| Where in the doc | Planted error | Debrief card |
|---|---|---|
| Section 2 ref (a) | "MCO 1000.6A" — the order exists, but the revision letter is wrong. AI invents revision letters in the right format. Looking up the order isn't enough; verify the revision. | Slide 34 · "What's wrong" · bullet 1 |
| Section 4, step 1 | "CMC Form 4790/142 (In-Processing Worksheet)" — plausible but invented form number. AI loves to invent CMC form numbers; the 4790 series is real but this specific form is fabricated. | Slide 34 · "What's wrong" · bullet 2 |
| Section 4, steps 1 & 5 | "Bldg 1284, Rm 203" and "Indoctrination Brief … every Monday at 0800 in Bldg 1284, Rm 110" — specific times and locations that look real but were never grounded in any source the model was given. Common AI confabulation. | Slide 34 · "What's wrong" · bullet 3 |
| Section 4 (omission) | Procedural gap. No mention of the medical, dental, or finance check-in stops most units actually require. This is the most dangerous category — you can't catch it by looking up a reference, only by knowing how the process really works. | Slide 34 · "What's wrong" · bullet 4 |
What's right (don't strike it): SOP structure and numbering, the effective-date and references section, the 72-hour reporting window. The form is a usable shell once you replace the invented references and add the missing stops.
Doc 3 errors — mapped to Slide 35
Open the document: Doc 3 · Training event after-action. Open the debrief slide: Week 1 deck · Slide 35.
| Where in the doc | Planted error | Debrief card |
|---|---|---|
| Section 3, para 1 | "98.7% qualification rate" — statistic to one decimal place suggesting precision the source data can't support. AI defaults to 98.7% over about 99% because it sounds more authoritative. | Slide 35 · "What's wrong" · bullet 1 |
| Section 3, para 2 | Inconsistent attribution. The document credits the S-4 communications detachment for net establishment that the S-6 actually executed. Para 2 names S-6 as the coordinator but then assigns execution to an "S-4 communications detachment" — there is no such detachment in the S-4. Wrong billet credited. | Slide 35 · "What's wrong" · bullet 2 |
| Section 3, para 1 | "The highest in the regiment for this fiscal year" and "the first time in five years" — positive but unverifiable claims. Awards-board territory. | Slide 35 · "What's wrong" · bullet 3 |
| Sections 4 & 5 | The recommendations parrot the observations. Compare the three Sustain bullets to the three Improve / Recommendations bullets — they're the same statements rewritten in the imperative. No real analysis underneath. | Slide 35 · "What's wrong" · bullet 4 |
What's right (don't strike it): Five-paragraph after-action structure, descriptive (not adversarial) tone, useful as a starting structure once the invented specifics and the parroted recommendations are replaced.
Closing the loop
After Slide 35, run the room into Slide 36 (the Verification Hierarchy). The whole point of the exercise is the pattern: confident references, false precision, procedural gaps, parroted recommendations. Match review depth to the cost of being wrong — a personnel package is always Centaur mode.