Project 5 — Deck in an Hour

“but test everything; hold fast what is good.” — 1 Thessalonians 5:21 (ESV)

Chapter: 5 — Choosing the Right Tool for the Job Due: End of Week 5 Submit: A link to a shared folder or a single combined PDF (Google Drive, OneDrive, or a PDF portfolio) containing your filled worksheet, your first-draft deck, your line-by-line critique, your corrected final deck, and your reflection. See Appendix A for account setup and submission mechanics. No GitHub, no code — these are ordinary documents and screenshots. Allowed tools: Any AI slide/presentation tool of your choice (e.g. Gamma, Microsoft 365 Copilot in PowerPoint, Gemini in Google Slides, Canva — see Appendix B); a general assistant and a research tool if your stack uses them; and your own two eyes for the verification, which is the graded part. AI policy: AI is used and verified, not trusted. The tool builds the deck; you own the verdict on every number and claim. Before you paste, check the Privacy gate (Appendix C) — this brief is safe, but the habit is graded. Every deliverable ends with an honest AI usage note.

The Setup

You are Maria Alvarez, operations coordinator at the Riverbend Regional Food Bank. You are not a designer, not a data analyst, and definitely not a programmer — you are the person who keeps the trucks loaded and the pantry stocked, and tonight you also have to build a board presentation.

The quarterly board meeting is tomorrow at 7 p.m. Nine volunteer board members — the people who fund the work and approve the budget — will sit in a church basement and decide, among other things, whether to release $18,500 for a second walk-in cooler you badly need. You have exactly one hour of quiet after the pantry closes to build the deck, and then you have to go home. Last quarter you stayed up until midnight fighting with slide layouts. Not tonight.

You have a clean one-page brief already written — every real number, the honest soft spot, the ask. This week you learned that a slides task belongs to a slide tool, not a chatbot, and that a good slide tool can turn a brief into a themed deck in minutes. So the plan is: let the tool build the deck in the first half hour, and spend the second half hour doing the one thing the tool cannot — making sure it tells the truth. Because you are the one standing in front of the board, and the board is trusting you, not the tool.

Setup (the starter)

This chapter ships two artifacts in code/:

• code/deck-brief.txt — the full one-page board-update brief: audience, purpose, tone, every real number, the honest soft spot (volunteer hours fell 11%), the ask ($18,500 cooler), and a note-to-presenter section the AI will not see. This is what you feed the slide tool.
• code/choose-your-tool.txt — the task → category → tool decision worksheet with the five-question method (Fit, Cost, Privacy, Quality, Learning curve) and the Privacy gate. You fill this in to choose and justify your tool.

Read the brief the way a board member will — quickly, and trusting every word. Then remember that the AI slide tool will not read it that way. It will helpfully “improve” your numbers, round your 8,410 households into a vaguer “8,000+,” and may add a confident statistic you never gave it. Your job is to catch every bit of that.

Learning Targets

You will demonstrate that you can:

• Choose a tool by evidence, not habit — pick a slide tool and justify it with the five-question method on the worksheet.
• Use a specialist well — generate a genuinely usable board deck from a brief in minutes.
• Verify polished AI output — catch fabricated statistics, drifted numbers, buried bad news, and missing asks by checking the deck line-by-line against the source.
• Own the verdict — ship a corrected final deck that tells the truth, with your name on it as the accountable human.

Normal Tier

Goal: Choose a slide tool with a documented rationale, generate a board deck from the brief, critique it line-by-line against the brief, and deliver a corrected final deck that tells the truth.

Required features

1. Filled worksheet (choose-your-tool — .txt or .docx). Complete code/choose-your-tool.txt for this task: name the task, confirm the category, score 2–3 candidate slide tools on Fit/Cost/Privacy/Quality/Learning, pass the Privacy gate, and record your chosen tool with a one-line reason it beat the runner-up.
2. First-draft deck (deck-first-draft — PDF export or screenshots of every slide). Feed the entire brief to your chosen tool and export/screenshot the result unedited. This is your “before” evidence — do not fix it yet. Note the tool and (if shown) the model, e.g. “Gamma, mid-2026.”
3. Line-by-line critique (critique — .docx or .xlsx). Compare the first-draft deck to the brief. For each problem, record: the slide, what the deck says, what the brief actually says, and the class of error — fabricated (a claim/number never in the brief), drifted (a real number changed or vaguened), buried (the honest soft spot hidden or dropped), or missing (a required element like the ask left out). Find at least the number drift, the honest soft spot, and the ask.
4. Corrected final deck (deck-final — PDF). Fix everything your critique found: every number matches the brief exactly, every fabricated claim is cut, the volunteer-hours soft spot is present and honest, and the $18,500 ask is clear. 6–8 slides. This is the deck Maria would actually present.
5. Reflection (reflection — .docx, ~400–600 words). Which tool you chose and why; the single most dangerous thing the AI got wrong and how you caught it; how long the generate step took versus the verify step; and an honest AI usage note. State plainly where the tool helped and where it nearly cost you.

Normal-tier rubric (out of 100)

Medium Tier (+up to 25% extra credit)

Test a second tool, the right way.

1. Second tool, same brief. Generate the deck a second time with a different slide tool, saved unedited as its own “before.”
2. Head-to-head on the method. Score both tools on the full five-question worksheet and write a short comparison: which won on Fit, which on Quality, which on Learning curve, which hallucinated more, and which you would actually keep. Treat any price as a mid-2026 snapshot and cite the vendor page.
3. Stack it. Add one more specialist to your workflow — e.g. a research/answer engine to confirm the county food-insecurity context, or an image tool for one clean header — and document the relay as a numbered stack (§5.8), naming the human step at the end.

Hard Tier (+up to 25% additional extra credit)

This is the judgment no tool can make for you, and it is graded as such.

Write MEMO (one page, .docx or .pdf), addressed to “the Riverbend leadership team,” recommending which single AI slide tool the food bank should standardize on for volunteer- and staff-built presentations — and defending it. You must:

• Name the tool, the runner-up, and the specific five-question evidence that decided it (not “it’s the best” — how it fit Riverbend’s actual needs: free/low-cost for a nonprofit, low learning curve for non-technical volunteers, privacy fit for donor data, export that works with what they already use).
• Name the failure mode you are accepting by standardizing on it, and the guardrail you would put around it (e.g. “every deck’s numbers are verified against source by a named human before it leaves the building”).
• Tie the recommendation explicitly to the spine rule — you choose the tool; you own the verdict — and to the discipline of test everything; hold fast what is good.

A memo that says “just use whatever” or “use the most famous one” fails. The grade is in the discrimination: showing you can weigh a real organization’s real constraints and defend one accountable line.

Submission

Put everything in one shared folder (Google Drive / OneDrive) or one combined PDF portfolio, and submit the link per Appendix A:

project-05/
  choose-your-tool.docx      (or .txt)   — filled worksheet + rationale
  deck-first-draft.pdf       (or screenshots of every slide, unedited)
  critique.docx              (or .xlsx)  — line-by-line, errors classed
  deck-final.pdf                          — the corrected deck you would present
  reflection.docx                         — 400–600 words + AI usage note
  MEMO.docx                               — Hard tier only

Do not submit code and do not use GitHub — these are ordinary documents. Redact nothing from the shipped brief (it is safe by design), but if you swap in your own real data, apply the Privacy gate first (Appendix C).

Hints (Read Before You Begin)

• Feed the whole brief, including the note-to-presenter section. The more context the tool has, the better the first draft — but it still will not know the things only Maria knows. That is why you verify.
• Save the “before” before you touch it. The unedited first draft is your evidence. If you fix as you go, you lose the comparison and the critique collapses.
• The three planted traps. The tool will almost always (a) drift or round at least one headline number, (b) be tempted to bury or soften the 11% volunteer-hours drop, and (c) risk dropping or vaguening the specific $18,500 ask. Check those three first, then hunt for a fabricated statistic.
• Numbers are checked against the source, not against what looks nice. 8,410 is 8,410. A fundraising slide that rounds it to “8,000+” is a slide that gave away 410 households you actually served.
• Time-box it like Maria. Thirty minutes to generate, thirty to verify and fix. The point of the project is that verification is fast when you have a source to check against — and non-negotiable.

What Mastery Looks Like (Beyond the Rubric)

A mastered submission is one where I can read your critique and final deck and tell that you were in the loop — that you caught the number that drifted, refused to let the tool bury the bad news, kept the ask sharp, and can say in one honest sentence where the tool saved you an hour and where it would have embarrassed you in front of the board. The pretty deck is table stakes; any tool makes a pretty deck. The verified deck — and your ability to defend a tool choice — is the project.

Coach’s Note — The temptation this week is to be so relieved the tool built a deck in minutes that you present it as-is. Resist it. The tool’s speed is real and its confidence is worthless. A board that catches one wrong number stops trusting all of them — and stops trusting you. The whole value you add tonight is the second half hour nobody sees.

When You’re Done (a short checklist)

• Worksheet filled; tool chosen with a five-question rationale; Privacy gate answered.
• First-draft deck saved unedited, tool named.
• Critique classes every error (fabricated / drifted / buried / missing).
• Every number in the final deck matches the brief exactly.
• The volunteer-hours soft spot is present and honest; the $18,500 ask is clear.
• Reflection names the most dangerous error and includes the AI usage note.
• (Hard) MEMO defends one standardized tool with real evidence and a named guardrail.

A theological footnote. “Test everything; hold fast what is good” (1 Thessalonians 5:21, ESV) is the entire project in eight words. The slide tool is neither to be worshiped nor despised — it is to be tested. You put its output to the proof against a source you trust, you hold fast to the good it gave you (a themed deck in minutes), and you let go of the rest without regret (the drifted number, the invented stat, the buried bad news). And there is a neighbor at the end of this: nine board members who will trust whatever number you put on the screen, and the families whose cooler depends on the board trusting you. To test the deck is to love that neighbor in the most practical way you can tonight — by refusing to let a machine’s untested confidence become their misplaced trust. The tool drafted. You tested. You held fast to what is good. You signed. That is the work.

See you next week.