scientific_poster_design v1.0.0

name: "scientific_poster_design" description: "Design a scientific or research poster that works at a distance: one claim per poster, column reading order, type sizes set for 1 to 2 metre viewing, figure and caption discipline, conference sizes (A0, A1, 36x48 and 48x36 inch), print checks, and a presenter pitch that matches the poster. Format neutral, works in PowerPoint, Slides, LaTeX, Illustrator, Canva, or HTML. Use when making, fixing, reviewing, or resizing a conference poster, poster session talk, or academic poster, or when a poster is a wall of text nobody reads."

Purpose

Turn a paper, thesis chapter, or project report into a poster a stranger can grasp while walking past, and a presenter can talk from for two minutes without reading it aloud. The poster has one job: carry one claim, prove it with figures, and start conversations. Detail belongs in the paper, a handout, or a QR linked PDF.

This skill is format neutral. It does not require LaTeX, PowerPoint, or any single tool, and it does not require generated images. Grids live in references/layout-grids.md, a full worked example lives in examples/worked-poster.md, and scripts/poster_check.py checks size, type, word count, contrast, image resolution, and print settings against the rules below.

Workflow

1. Get the event spec before designing anything

• Record the exact board size, orientation, and any required sections, logos, or judging criteria from the conference or course guidelines. The event spec beats every default in this skill.
• Common sizes, for when the spec allows a choice:
• A0: 841 by 1189 mm, 33.1 by 46.8 in, one square metre (ISO 216).
• A1: 594 by 841 mm, 23.4 by 33.1 in, half the area of A0, same 1 to root 2 shape.
• 48 by 36 in landscape (121.9 by 91.4 cm): a common US conference size.
• 36 by 48 in portrait: the same sheet turned upright.
• Set the design file to the final print size at the start. Designing small and scaling up makes every font size on the page a lie.
• Landscape keeps more of the poster near eye level. Portrait is fine when the board demands it, but keep must-read content out of the bottom quarter where readers have to stoop.

2. Write the one claim

• State the main finding in one plain sentence: what was compared or measured, what happened, and by how much if the study supports a number. If the user cannot write that sentence yet, the poster is not ready to lay out. Help them write it from their results section.
• Test the claim: could a passerby repeat it after reading it once, from two metres, in about five seconds? If not, shorten it. Remove jargon, remove hedges that hide the result, keep the one limit that changes how the claim should be read.
• Everything on the poster either supports the claim, explains how it was earned, or states its limits. Cut anything else, however interesting. A second study hiding in the same poster gets its own poster.

3. Budget the words and the space

• Word ceiling: 800 words total on the poster, with 600 to 800 as the working range in most university guides. One guide puts the floor test plainly: a reader should finish in about 3 to 5 minutes. Over the ceiling, cut text. Never shrink type to fit more words.
• Space split to aim for: roughly 20 percent text, 40 percent figures, 40 percent empty space. Empty space is a feature. It is what lets a crowd scan the poster without stopping shoulder to shoulder.
• Draft in this order: claim, figure list, headings, bullets, captions, methods, references last. Writing the introduction first produces a paper pasted on a board.

4. Lay out the reading order

• Use a grid from references/layout-grids.md: Grid A classic three column as the default, Grid B billboard centre (claim in the middle, silent presenter bar on one side, data bar on the other) for crowded general sessions, Grid C two column for portrait boards.
• Reading order runs top left to bottom right, down one column and then down the next. If the layout breaks that flow, number the sections or add clear arrows. Do not make readers hunt for step two.
• Title band across the full width: short title, authors, affiliations. Keep the title to two lines at most, and write it as a finding or a clear topic, not a thesis title copied whole.
• Keep each figure next to the text that cites it. Align columns on one edge each. Left align body text, keep paragraphs to five lines or fewer, and keep column lines around 45 to 65 characters.

5. Set type for the viewing distance

Minimums, drawn from university poster guides (Temple, Yale, University of South Florida, Adelphi):

• Title: at least 72pt. Working range 85 to 125pt, larger for A0.
• Authors: at least 48pt on a full size poster.
• Section headings: at least 36pt, 42pt or more is better.
• Body text: at least 24pt. That is a floor, not a target.
• Captions and references: at least 18pt. Axis labels on figures follow the caption floor, never smaller.

Distance rule: comfortable body size grows with how far back readers stand. A widely used table gives about 8pt at 0.35 m, 16pt at 0.7 m, and 32pt at 1.5 m, so plan on roughly 22pt per metre of viewing distance for comfort. At the usual 1 to 2 metre poster distance, that puts comfortable body text near 28 to 36pt, with letters around 1 cm high as a cross check. Budgets: if the board sits in a tight aisle, 24 to 28pt body can work. If a crowd or a rope keeps readers at 2 metres, design at 32pt and cut words to pay for it.

• Use one or two fonts, sans-serif for body text (Arial, Helvetica, Verdana, Calibri class). Keep the same role at the same size everywhere.
• Avoid all caps for anything longer than a short heading. It slows reading.
• Run python3 scripts/poster_check.py spec.json (or --example to see the format) before calling a layout done. It checks the minimums above, the distance rule, word count, layout split, contrast, image resolution, bleed, embedded fonts, and colour mode, and prints a fail, warn, ok line per check.

6. Figures and captions

• One hero figure carries the claim. Make it the largest element after the title and claim strip, big enough that its axes and labels read at the design distance.
• Every figure earns its place by answering one question. Merge panels that make the same point, drop figures that only show a method detail the bullets already state.
• Each figure gets a number, is cited in nearby text, and carries a caption that works alone: takeaway first in plain words, then what is plotted, sample size or conditions, and what a reader should notice. A reader who only reads captions should still get the story.
• Check effective resolution at the printed size, not the file's stated dpi: pixels wide divided by printed inches wide. At least 150 dpi at final size for distance viewing, 300 dpi for anything readers will inspect up close. Redraw charts as vectors where the tool allows it, they stay sharp at any size.
• Do not encode meaning by colour alone. Pair colour with labels, line styles, or direct annotations, and check chart colours for contrast against the background (3 to 1 for meaningful graphics under WCAG 2.x AA).

7. Colour and accessibility

• Light background, dark text is the safe default for a lit poster hall. Whatever the palette, body text needs a contrast ratio of at least 4.5 to 1 against its background, and large headings at least 3 to 1, per WCAG 2.x AA. The checker computes this from hex values.
• Two or three colours with jobs: one for structure (title band, headings), one for the data highlight, and neutral text. Colour that decorates competes with colour that means something.
• Proof colours on paper if the stakes justify it. Screens are backlit and print is not, so saturated screen colours print darker and duller than they look.

8. Print checks

• Export a PDF at final size. Fonts embedded. Colour converted to CMYK or soft proofed, so print colour does not shift without warning.
• Bleed: 3 mm on all sides is the common spec (0.125 in in US shops). Confirm the printer's own number, large format shops sometimes want 5 mm or more, and keep text and logos inside the safe zone away from the trim edge.
• Resolution: 150 dpi at final size is a working floor for large posters viewed at 1 to 2 metres, 300 dpi for close viewing and fine detail. Upscaling a small image does not add detail. Replace it or print it smaller.
• Do the cheap proof: print one column or the hero figure at 100 percent on office paper, tape it to a wall, and read it from the design distance. Catch small type and muddy figures here, not on the full price print.
• Do the five second test: show the full poster (projected or printed draft) for five seconds, look away, and state the claim. If you cannot, the claim strip and hero figure are not doing their jobs yet.
• Travel check: confirm tube or folder size against the poster size, and carry a PDF copy plus the talk outline on a phone in case the print is lost or damaged.

9. Write the pitch to match the poster

• Prepare three lengths, all starting from the printed claim, word for word:
• 10 seconds for the passerby: the claim, nothing else.
• 60 seconds: question, method in two sentences, hero figure walked in reading order, the durability or follow up result, one limit stated up front.
• Deep version for the expert: the extra data in the side bar or handout, limits in full, and what the next study would change.
• Rehearse pointing, not reading. The presenter faces the visitor, gestures at figures, and lets the poster carry the wording.
• End every pitch the same way: point to the QR code linking to the paper or full poster PDF, state one thing available there that is not on the poster, and give a contact route.
• Anticipate the three hardest questions (weakest method choice, strongest alternative explanation, limit that matters most) and put the honest answer for each in the deep version notes, not on the poster face.

Output Contract

For a design or redesign task, produce:

1. The event spec as recorded (size, orientation, required sections), or a clear statement of the default size chosen and why.
2. The one claim, in one sentence, plus the title built from it.
3. A word and space budget: target word count and the text, figures, whitespace split.
4. A layout plan naming the grid used, the reading order, and what sits in each column or bar.
5. A type table: title, authors, headings, body, captions, each with its size and the viewing distance it was set for.
6. A figure list: each figure, the question it answers, its printed size, and its caption written takeaway first.
7. The poster_check.py output for the poster's spec, with every fail fixed or explicitly flagged to the user, and warnings explained.
8. Print settings: final size, bleed, colour mode, resolution at final size, file format.
9. The pitch: 10 second, 60 second, and deep versions, plus the three hard questions with honest answers noted.

For a review task, produce the same checks as a verdict list (pass, warn, fail per item) and name the smallest set of changes that fixes every fail, starting with claim, type size, and word count before any colour or styling changes.

Operating Rules

1. One claim per poster. Two claims means two posters, or one claim demoted to a supporting result.
2. Never shrink type below the minimums to fit content. Cut content. Small type is how posters fail.
3. Never invent results, numbers, sample sizes, or citations to fill a panel. If a panel has no real content, the layout is wrong, not the research. Flag the gap to the user.
4. The conference spec overrides every default here. A judged poster follows the judging rubric layout expectations, even when a billboard layout would read better for a general session. State the trade off when it matters.
5. Figures carry the argument, text explains the figures. If a draft reads fine with the figures removed, it is a paper summary, not a poster.
6. Verify every size, ratio, and standard you cite from the references in this skill or a primary source. Poster advice varies by institution, so present the minimums as guide values with a range, not as a universal law.
7. Worked example data in examples/worked-poster.md is illustrative layout practice material. Never present it as a real study or cite its numbers.