Executive artifacts lead with the conclusion, then support it — never build to a reveal. Barbara Minto’s pyramid (The Minto Pyramid Principle): one governing thought at the top, supported by 3–4 grouped arguments, each supported by evidence. The military version is BLUF — bottom line up front.
The governing thought is a claim, not a topic: “Churn is concentrated in month-2 self-serve users; fixing onboarding recovers ~$2M ARR” — not “Churn analysis.”
Each section answers the question its parent raises. If a section doesn’t support the governing thought, it goes to the appendix or gets cut.
Order groups by MECE logic (mutually exclusive, collectively exhaustive) and by what the audience will ask next.
This inverts narrative instinct — journalists and novelists build tension; executives triage. Use the three-act arc within sections and for standalone stories; use the pyramid for anything a decision-maker reads.
Before outlining, write one complete sentence that (a) states your point of view, (b) conveys what’s at stake, (c) is a full sentence (Knaflic’s test). If you can’t, the analysis isn’t finished — go back to the data. This sentence becomes the title of an exec summary, the headline of slide 1, or the hook of an infographic.
One page. Answer first, three to four supporting points, each with its number and its “so what.”
Every number carries context (vs. what, over what period, why it matters) — a bare KPI is not a finding.
The cold-reader test: someone with no context extracts the decision and the ask in 30 seconds. Test this literally (hand it to a fresh reader — or a fresh model instance — and ask “what’s the decision?”).
State confidence and caveats in one honest line, not a hedge-everything paragraph.
Deliver the design layer, not just the content: even when the deliverable is markdown, specify the visual hierarchy — headline treatment, which numbers become big-number callouts, what gets the emphasis color, table styling — so whoever produces the final artifact inherits the design intent. Content without layout guidance is half the deliverable (page-design.md has the system to draw from).
Decide which you’re making (Duarte’s distinction):
Presentation slides (spoken over): one idea per slide, headline states the claim (“assertion-evidence” style: title = assertion, body = the evidence chart), minimal text, big type. The presenter is the narration.
Slidedoc (read unaccompanied, sent around): denser, self-explanatory, full sentences allowed, annotations carry the narration. Most “decks” that get emailed are slidedocs pretending to be presentations — design for how it will actually be consumed.
Slide titles across the deck should read as a coherent argument on their own (flip through titles only — is the story there?).
See dreambase-visualization-design‘s references/infographics.md — one message, three-act structure, 5-second rule. This skill adds the layout/typography layer on top of those rules.
The same analysis often ships as multiple artifacts. Design them as one system, not three documents:
Exec tier: the decision, the three numbers that matter, the ask (one page / three slides).
Manager tier: + drivers, comparisons, and what changes operationally.
Analyst tier: + method, data lineage, uncertainty, reproducibility.
Cut down from the full story; never pad up from a summary. Keep the same headline claim, palette, chart forms, and terminology across tiers so the artifacts corroborate each other.
Inherited from dreambase-visualization-design and non-negotiable in narrative formats, where the temptation is strongest: what the data shows (observation), what plausibly explains it (interpretation), what to do (recommendation) — visibly separated, never blended into one confident sentence.