RaterZero · Critiques

Layout is a decoration filter.

I judge visual work for a living: AI-generated slide decks and images, and computer vision training data. These studies show how I read a slide. In each one, the facts were on the page. The layout decided whether they reached the person who had to act on them.

These are independent studies of publicly shared material, not client work. Each "before" panel rebuilds the original slide's structure and key wording so the problems can be marked up. Follow the source link to see the original itself. Each "after" panel is my redesign.

The five checks I run on every slide

  1. H · HEADLINEThe title states the findingA reader who sees only the title should get the point, not just the topic.
  2. R · RANKVisual weight matches importanceThe most important fact gets the most size, contrast and position.
  3. E · EVIDENCENumbers come with scaleA figure needs its baseline, its unit and its comparison to mean anything.
  4. D · DECISIONThe ask is framed up frontIf the audience must choose, they see the choice before they read the detail.
  5. S · SIGNALColor and type carry meaningEvery accent, icon and weight change should tell the reader something.

Study · Engineering briefing · 2003

The warning that sat in the smallest type

Boeing's January 2003 debris assessment for NASA after the foam strike on Space Shuttle Columbia (STS-107). Source: archived NASA copy of the briefing (PDF).

HRE
Before

Reconstruction. Bullet levels follow the original; some middle bullets are shortened.

  1. 1
    The title argues against its own data"Conservatism" tells a manager the model errs on the safe side. The slide's last bullet says the model doesn't apply to this case at all.
  2. 2
    The finding that matters is buried"Outside of test database" sits two levels down. Nested bullets rank ideas by indentation, so the decisive point looks like a footnote.
  3. 3
    A 640× gap reads as a minor detail1,920 cubic inches against 3 is a 640-fold jump past anything tested. It appears as a raw comparison in the smallest type on the slide, with no ratio and no consequence.
  4. 4
    One word stretched over four very different sizes"Significant" describes a modelling error, an energy threshold, tile damage and a gap far outside the data. It stops telling the reader how big anything is.
After

What changed

The headline now carries the finding the original buried. The two volumes are drawn to one scale, so the gap is something you see, not something you read. The test bar is barely a hairline, and that is the point.

The supporting logic shrinks to three short columns in reading order. The last column, the one a decision depends on, gets the accent color.

Study · City budget deck · 2025

A ten-slide decision with the question at the end

City of Burleson, Texas: FY 2025-2026 Budget & Capital Improvement Program Overview, presented to City Council on May 27, 2025 (about 81 slides). Source: council meeting attachment (PDF). This study covers slides 20 to 29, the General Government CIP.

HDR
Before

Reconstruction of slide 21 with titles of the slides around it. Red outlines mark the five CIP tables.

  1. 1
    The title names a topic, not the finding"Tax Implications" says nothing. The real message is in the first bullet: keeping the plan whole needs a $0.0547 rate increase. Most of the deck's titles work the same way.
  2. 2
    Council has to compare five tables by eyeCurrent, redlined and proposed versions of the same CIP appear at two tax rates. What changed between them lives in redline marks and on slide 24, not in one view.
  3. 3
    The decision arrives lastThe questions council must answer come on slide 29, after all the evidence. Members read nine slides without knowing what they are being asked to weigh.
After

What changed

Ten slides become one decision slide that council sees first. The title states the trade-off, and the two paths sit side by side with what each one costs.

The committee's pick gets the only accent on the slide. The five CIP tables move to an appendix, where they back up the decision instead of standing in for it.

Study · AI-generated slide patterns

Four equal cards and no point

A composite I built to show failure patterns that come up again and again in AI-generated decks. The company, Harbor & Pine Coffee, and all of its numbers are made up for this study.

HRES
Before

Composite slide, made for this study.

  1. 1
    The title is a label with an emoji"Q3 Highlights" could head any deck. The rocket adds mood, not information.
  2. 2
    The subtitle is filler in low contrastHalf-transparent white on a light gradient is hard to read, and "driving synergy" says nothing a reader could check.
  3. 3
    Numbers without baselines300% growth of what, since when? Four stats get identical cards, so nothing tells the reader which one matters or how they connect.
  4. 4
    Text clipped mid-sentence, one card out of lineThe only sentence with real content overflows its fixed-height card and is cut off. That card also sits lower than its neighbours.
  5. 5
    A chart drawn for decorationNo axis, no units, no labels. It suggests "up and to the right" without showing any data.
After

What changed

The headline says what happened and why, so the slide makes one argument. The chart shows each channel against last quarter on one scale, and only the channel that explains the growth gets color.

The "300%" figure had no stated base, so it's gone. The partner count survives because it explains the wholesale jump. The rating moves to the appendix because it belongs to a different story.

How these were made

The before panels are rebuilt in HTML from the published originals, keeping their structure and key wording. They are not images of the originals. The redesigns use only facts that appear in each source, except in the third study, where all data is fictional. I have no affiliation with NASA, Boeing or the City of Burleson.