---
title: "[Your Title Here]"
subtitle: "Final Project · STAT 540 · Computing in Statistics"
author: "Your Name"
institute: "Department of Statistics, University of South Carolina"
date: today
format:
  beamer:
    theme: Madrid
    colortheme: beaver
    fonttheme: structurebold
    aspectratio: 169
    navigation: horizontal
    header-includes: |
      \usepackage{booktabs}
      \setbeamertemplate{footline}[frame number]
      \setbeamerfont{footnote}{size=\tiny}
      \usepackage{fvextra}
      \DefineVerbatimEnvironment{Highlighting}{Verbatim}{breaklines,breakanywhere,commandchars=\\\{\}}
      \fvset{fontsize=\footnotesize}
      \makeatletter
      \renewcommand{\verbatim@font}{\ttfamily\footnotesize}
      \makeatother
execute:
  echo: false
  warning: false
  message: false
---

```{r}
#| include: false
# =============================================================================
#  READ THIS FIRST — THEN LEAVE IT ALONE. It never appears in the PDF.
#
#  Every instruction in this template lives inside a comment: visible to you
#  in the editor, invisible in the rendered deck. There is nothing for you to
#  remember to delete.
#
#  FIVE MINUTES. THREE CONTENT SLIDES. HARD STOP.
#
#  The three slides are fixed:
#      1. The analysis and the ask
#      2. What verification found
#      3. What it means
#
#  `echo: false` is set in the YAML above, so your code will not appear on the
#  slides. That is deliberate — nobody can read code from the back of a room.
#  The code lives in your report.
#
#  THIS IS THE R VERSION (knitr engine — no Jupyter needed).
#  If your own analysis is in Python, start from project_slides_template.qmd.
#
#  TO BUILD THE PDF:   quarto render project_slides_template_R.qmd
#  That is the whole process — Beamer produces a PDF directly. There is no
#  browser step and nothing to print.
#
#  SETUP CHECK:        quarto check knitr
#  You also need LaTeX. If `quarto render` complains that pdflatex is missing,
#  run:  quarto install tinytex
#
#  NOTE FOR EDITING: do not put a comment block above the first "##" heading.
#  Beamer wraps anything before the first slide title in a frame of its own,
#  and you get a blank slide. That is why this block lives inside a chunk.
# =============================================================================

# knitr writes PDF figures for Beamer by default, so plots stay sharp when
# projected. These sizes keep a figure and three bullets on one slide.
knitr::opts_chunk$set(fig.width = 5.4, fig.height = 1.9, dev = "pdf")
options(width = 70)
```

## The analysis and the ask

<!--
  THIRTY SECONDS ON THE ANALYSIS, THIRTY ON THE ASK.

  The temptation is to spend three minutes explaining your dataset. Resist it —
  the analysis is the setup, not the story.

  The audience needs: what the data are, what you asked of them, which
  direction you translated, and what you told the assistant to do.
-->

- **Dataset:** … *(what it is, who it describes, one row)*
- **My question:** … *(one sentence)*
- **I wrote it in:** … $\rightarrow$ **the assistant translated to:** …
- **Assistant and date:** …
- **The risky elements I built in:** … *(two or three, briefly)*

## What verification found

```{r}
#| label: headline-figure
#| fig-align: center
# The single most useful visual: the comparison table as a figure, the
# divergence plotted, or your adversarial case. One thing, not three.

# ==== DELETE THIS PLACEHOLDER and draw your own figure ====================
par(mar = c(2, 2, 1, 1))
plot.new()
text(0.5, 0.5, "your one figure goes here", col = "grey60")
box()
# ==========================================================================
```

<!--
  ONE FINDING, SAID WITH A NUMBER. Either the most interesting divergence, or —
  if everything matched — your adversarial case.

  Say the mechanism out loud:
     "The translation used sd() where mine used np.std(), so every standard
      deviation came back 1.2% too large"
  beats "there were some differences."

  If your most interesting finding was TIER 2 — a denominator, an unweighted
  average, an assumption nobody checked — this is where it goes, and it is
  usually the better talk. Both programs agreeing on the wrong answer is a more
  surprising story than two programs disagreeing.

  If the error turned out to be in YOUR OWN analysis, say so. That is the more
  honest talk and it is graded as such.
-->

- **What I compared:** …
- **What diverged (or what didn't):** …
- **The mechanism:** …

## What it means

<!--
  ONE SPECIFIC REFLEX you will carry forward. Not "be careful with AI" —
  something you would actually type.

     Good: "I now check row counts before and after every filter, in both
            versions."
     Bad:  "AI output should always be verified."

  Leave room. Two minutes of questions follow, and a wall of text behind you is
  a distraction while you answer. Have your report open — you will be asked
  where a number came from.
-->

- **What the assistant got right:** …
- **What it got wrong, and what kind of error that is:** …
- **What I will check by reflex from now on:** …

## Questions?

\begin{center}
\Huge Questions?
\end{center}

<!--
================================================================================
  WHAT NOT TO PUT ON THESE SLIDES

  - THE TRANSCRIPT. Nobody can read it from the back of the room, and it is in
    your report.
  - ALL SEVEN TIER 2 CHECKS. Pick the one that changed something.
  - A TABLE OF NINE NUMBERS THAT ALL MATCHED. If everything matched, the
    adversarial case is your slide — it is more interesting than the agreement.
  - CODE. `echo: false` is set for a reason.

  ==========================================================================
  A FIGURE MADE IN THE OTHER LANGUAGE

  If your headline visual came from the translated program rather than from
  R, save it as a .png or .pdf from that program and put it on the slide
  with a plain Markdown image, replacing the code chunk above:

      ![](my_figure.pdf)

  A PDF or SVG will stay sharp when projected; a low-resolution PNG will not.

  ==========================================================================
  IF THE RENDER FAILS

  "there is no package called ..." — install it in R first. This template
  itself needs no packages beyond base R and rmarkdown.

  "pdflatex not found" — run `quarto install tinytex` once, then render again.
================================================================================
-->