best-practices-slide-design

# Slide Design Best Practices Craft rules for turning claim-honest deck drafts into persuasive slides. Every rule cites exemplar slides rendered from the human's REAL decks (`references/exemplars.yaml`, images in `assets/`). Design is ADVISORY; claims are LAW — nothing here overrides the pitchdeck compiler's gates. ## The three principles (from the deck author, 2026-08-06) 1. **Distance legibility drives chrome.** The header band exists so header and body separate at 5–20 ft. If a slide's title zone is not distinguishable at thumbnail scale, it fails. (`cybersummit-04`) 2. **Animation is rhetorical.** Builds reveal in the order the argument unfolds; motion with no argumentative role is noise. Duplicated-slide runs that fake builds become `ContentReveal.STEP` fragments. (`cybersummit-42..45`, anti-exemplar) 3. **Multi-channel reinforcement.** Headline asserts it, chevrons state it, the diagram shows it, the metaphor badge sets the emotional register — the SAME idea on every channel. Text-only slides get a reinforcement proposal; text and visual making different points is a failure. (`cybersummit-18`) ## The archetype catalog `references/DESIGN_SLIDES.md` classifies ALL 263 corpus slides into ten archetypes (assertion+art 30%, section-divider 20%, bullets, art-rich, dense-reference, mixed Q&A, art-only, close, toc, cover) with measured geometry, exemplars, cross-archetype design laws, and the compiler mapping or gap for each. Pick the archetype before composing; the machine-readable assignment lives in outputs/house-slides/archetypes.json. ## Slide-level rules | Rule | Statement | Exemplar | |------|-----------|----------| | headline-as-assertion | Titles are takeaways ("LLMs are Expensive"), never labels ("Cost Analysis") | cybersummit-49 | | thesis-as-statement | Thesis = ONE hero-size assertion (64–112pt) + one icon, qualifiers in notes/footer | reqml-12 | | density-5x5 | ≤5 words/line, ≤5 lines, ≤4 takeaways before a visual; median ~16–33 words/slide | anti: cybersummit-21 (224 words) | | one-big-diagram | Process/how slides earn ONE large diagram with labeled endpoints, not bullet paragraphs | cybersummit-12, cybersummit-18 | | one-idea-per-slide | Channels must agree on a single idea; split slides that argue two things | cybersummit-18 | | builds-are-fragments | Reveal order = narrative order, via fragments, never duplicated slides | anti: cybersummit-42..45 | | cover-brand | Cover = wordmark + one-phrase tagline + brand glyph, ~5 words | cybersummit-01 | | recap-device | Long decks restate the core assertion mid-deck ("What's the point, again?") | reqml-12 | ## Fonts and styling (house profile) Measured across 6 real decks (2023–2026, all audiences) — see `/mnt/storage12tb/skills/pitchdeck/outputs/style-references/` for per-deck profiles and `house-style-synthesis.yaml` for the corpus synthesis. - **Type**: Calibri (humanist sans) everywhere; Consolas for code; Roboto in diagrams. Scale: 64pt hero (112pt statement slides) / 36 section / 28 title / 24 lead / 20 body / 16 support / 12 caption. - **Color**: petrol `#065E7C` primary (brand constant across every deck); ink `#292929` (never pure black in body); warm `#D39500`/gold `#D6A300` and green `#6F8E30` as inline emphasis; program-blue `#26558E` joins for DARPA/PI audiences; red sparingly (`#A14240`). - **Chrome**: full-width teal header band with white title; circular line-icon metaphor badge top-right; footer = distribution statement + page number (+ sponsor strip on public decks). - **Bullets**: teal chevron `>` level 1, small square level 2, em-dash level 3; key words emphasized inline via underline or warm color, and color-coded label prefixes (**Problems/Goal/Solution/Impact**). ## Theme templates (`themes/*.json`, `pitchdeck.theme_template.v1`) Drop-in themes for /pitchdeck. `theme_tokens` maps directly onto today's `ThemeTokens` (accent, heading_font, body_font); the full palette, type scale, chrome, and density budgets are staged for the extended tokens (#1262). Audience is a PARAMETER, not a new design: | Template | Audience | What changes | |----------|----------|--------------| | sparta-house-conference | conference talks | humor devices, metaphor badges, big icon diagrams | | sparta-house-sbir | SBIR / investors | value prop before ToC, gold up-weighted, humor down | | sparta-house-program-review | DARPA / PI meetings | program-blue, early pipeline-position slide, accomplishments section, notice slide, recap devices | ## Deck arc Decks are ASSEMBLED from reusable modules, not drafted from scratch: the ACERT mini-arc (Origin → Problem-Solution → How → Journey) appears nearly verbatim in four decks. Standard arc: cover → ToC → value prop → problem/solution → vision → product mini-arcs → roadmap → discussion, with a Boneyard appendix absorbing overflow. Audience templates reorder this arc (`arc_overrides`), never invent a new one. ## How to apply - **Planning**: pick the audience template; draft the narrative arc first; slot claim-bound modules into it; every slide gets an assertion headline bound to spans (NUMERIC_UNBOUND keeps "70% faster" honest). - **Critique**: score each slide against the rules table; emit advisory `DESIGN_*` findings with the violated rule + exemplar; propose fixes as simulate-validated EditProposals (layout changes stay governance-gated). - **New style reference**: analyze a PPTX with python-pptx (fonts, colors, sizes, words/slide, shape census) + render representative slides; write a `pitchdeck.style_reference.v1` YAML next to the existing profiles.