Red Pen Review

Reviewer Worksheet (Markdown)

The printable companion. One copy per document under review. Captures the six-question checklist, a flagged-claims log, the “what's right” column so you don't throw the baby out, and a sign-off block. Open in any text editor or render in your favourite Markdown previewer.

The six-question checklist

This is the on-slide checklist students apply during the 10-minute individual mark-up. Project it on Slide 31 while they read.

1. References & citations. Every cited order, MARADMIN, form number, study, URL, or page reference — does it actually exist? Cited MARADMINs are a common AI fabrication.
2. Facts & statistics. Every number, percentage, dollar figure, headcount, or date — can you verify it in a source you control? AI loves false precision: 98.7% sounds verifiable; it usually isn't.
3. Internal consistency. Do names, ranks, dates, units, and units of measure agree everywhere they appear? Inconsistent attribution is a classic AI tell.
4. Procedural accuracy. Do the steps, sequence, and approval chain match how the process actually works in your unit? Procedural gaps are the most dangerous category — they require you to know how the process really works.
5. Substance vs. style. Strip out filler vocabulary — tireless, unparalleled, selfless, robust. Is there real content underneath, or only polished filler?
6. The signature test. Would you sign it and put it in front of your supervisor / commander / the awards board? If not, mark it.

Rule of thumb: finding 3–4 issues per document is doing well. Finding 1 or 0 is the lesson — go back over the references and the numbers a second time.

The three sample documents

A vetted, UNCLASSIFIED set of three sample documents lives on the site so you don't have to draft them from scratch the first time you run Module 4. Drop them in the Microsoft Teams chat at Slide 28 (break) and again at Slide 31 (exercise). The briefs below describe each document's shape and the planted errors — the answer key maps every error back to its debrief card on Slides 33–35.

Why we ship them. Module 4 is the highest-stakes 25 minutes of Course 1. New instructors used to draft three documents from scratch the first time they ran it; the shipped set gives every cohort a known-good starting point and lets students who missed the live session re-run the exercise on their own.

What's in each document — the briefs

The briefs below are the recipe behind each shipped sample. They're useful when you want to swap in a unit-specific variant per cohort — the answer key tells you which planted errors each new variant still needs to contain.

How to use the shipped samples: drop the three URLs above in the Microsoft Teams chat at Slide 28 (break) and again at Slide 31 (exercise) so latecomers have them, then run the debrief from Slides 33–35. The instructor answer key tells you which planted error each card on each slide is revealing.

If you'd rather stage your own: draft each document with M365 Copilot, Azure OpenAI, or any approved tool, using the briefs above as the recipe. Give the model thin context on purpose — that's how it produces the kind of organic errors students need to learn to spot. Save the three documents as PDFs or shared-doc links and drop them in chat alongside (or instead of) the shipped samples.

The Verification Hierarchy

Not every document deserves the Red Pen Review. Match review depth to the cost of being wrong.

• High-stakes documents (legal, personnel, financial) — line-by-line verification of every fact, reference, and number. Centaur mode. Human verifies everything. A personnel package is always here.
• Medium-stakes documents (correspondence, reports, briefings) — spot-check key claims, verify references, check tone. Mixed mode.
• Low-stakes documents (first drafts, brainstorming, formatting) — quick review for obvious errors. Cyborg mode.

What to surface in the debrief

Walk the room one document at a time. Take a count before revealing the cards on each debrief slide.

• Doc 1. Ask: “Who flagged the percentage? Who flagged the cited MARADMIN?” The made-up reference is the dangerous one — AI invents reference numbers in confident, plausible formats. Sign one and you've signed a fabrication.
• Doc 2. Ask: “Who looked up the form number? Who flagged the procedural gaps?” Most rooms catch one but not both. Procedural gaps are the most dangerous category — there is no substitute for domain knowledge.
• Doc 3. Ask: “Who caught the wrong attribution? Who flagged the false precision?” AI defaults to 98.7% over about 99% because it sounds more authoritative — even when there is no data behind it.