---
name: designing-presentations
description: >-
  Designs high-stakes presentations as engineered cognition change, not information transfer — for
  researchers, founders, and engineers giving talks, pitches, defenses, exec readouts, and demos. Use
  when planning or critiquing a talk/deck/pitch, choosing an order (pyramid, SCQA/SCR, BLUF, strategic
  narrative), writing a governing sentence, designing assertion-evidence slides, building a
  Q&A/objection plan, or sanity-checking a "science of presentations" claim. Triggers — slide deck,
  pitch deck, keynote, conference talk, thesis defense, exec/board readout, demo, "make my slides
  better", "how should I open/close", "is this presentation myth true", storytelling stats, 7-38-55,
  attention span, 10/20/30.
---

# Designing presentations

## Contents

- [The one idea](#the-one-idea)
- [The procedure (STEPs 0–9)](#the-procedure-steps-09)
- [The non-negotiables (apply at every STEP)](#the-non-negotiables-apply-at-every-step)
- [Execution model — design solo, audit fan-out](#execution-model--design-solo-audit-fan-out)
- [Pointers (depth lives in references/)](#pointers-depth-lives-in-references)

## The one idea

A presentation is **engineered cognition change, not information transfer.** The deliverable is never
"the audience now knows X" — it is "the audience now **decides / does / believes / is** X." Every move
below is subordinate to one quantity: the **cognitive gap** between where the audience *starts* and the
single state you need them to *reach*. Design backward from that gap. A talk that "covers the material"
with no target state is the master anti-pattern; it produces a data dump.

This skill is a **decision procedure**, not a tip list. It teaches durable judgment — what to do and
*why*, and the regime where each default **flips** — not delivery folklore Claude already knows
("make eye contact"). Contested empirical numbers, named studies, and debunked myths are quarantined
in references with citations; the body stays durable.

## The procedure (STEPs 0–9)

Run these in order. STEPs 0–4 are the **front half** (decide *what* must change before touching an
order or a slide); STEPs 5–9 build and deliver it.

| STEP | Move | Owns the answer to | Depth |
|---|---|---|---|
| **0** | **Pick the medium** — deck vs prose memo vs live demo — *before* content. Prose memo → hand off to `issuing-technical-memoranda` | "Should this even be a deck?" | `references/context-playbooks.md` |
| **1** | **Name ONE goal as a future audience STATE** | "After this, [audience] will [decide/do/be] X" | `references/audience-and-message.md` |
| **2** | **Model the audience's initial state** on five axes | knowledge · stakes · doubts · decision-criteria · expertise | `references/audience-and-message.md` |
| **3** | **Defeat the Curse of Knowledge** with concreteness | "Can a non-expert picture every load-bearing claim?" | `references/audience-and-message.md` |
| **4** | **Compute the cognitive gap** + the **governing sentence** | the ordered list of belief-shifts; the one-sentence thesis | `references/audience-and-message.md` |
| **5** | **Order** the belief-shifts by the audience's *next question* | pyramid / SCQA / SCR / BLUF / strategic narrative | `references/structure-and-narrative.md` |
| **6** | **Build slides** as assertion-title + visual evidence | one message per slide; the titles-only test | `references/slides-visuals-cognitive-load.md` |
| **7** | **Instantiate the per-context skeleton** | research talk · pitch · readout · defense · demo · keynote | `references/context-playbooks.md` |
| **8** | **Design open / close / Q&A** as structure, not improvisation | the explicit ask; the objection plan | `references/delivery-objections-and-evidence.md` |
| **9** | **Diagnose the binding constraint** (ethos/pathos/logos) and **compress** | spend scarce effort on the bottleneck, not uniformly | `references/delivery-objections-and-evidence.md` |

### STEP 0 — Pick the medium

If the artifact will be **read alone** rather than presented, a deck is the wrong medium — write a
prose memo (the reasoning survives without you). If the value *is* a live reveal, it is a demo, not a
slide walk. Choose the medium before the content. (Deck-vs-doc-vs-demo gate: `references/context-playbooks.md` §1.)

**The prose-memo branch exits this skill.** `issuing-technical-memoranda` owns it. That means the
cover block, the authority line, the addressee declaration, the release marking. This step decides
only the medium, never the memo's wrapper.

### STEP 1 — One goal as a future state

> *"After this, **[specific audience]** will **[decide / do / believe / be]** **X**."*

Reject information-frame goals ("explain the research," "cover the project"). One goal, phrased as a
*state change*. Three goals = three talks. (`references/audience-and-message.md` §2.)

### STEP 2 — Model the initial state (five axes)

Knowledge · interest/stakes · **doubts** · decision-criteria · expertise. Write them down for *this*
room; do not carry them in your head. The doubts axis becomes the Q&A objection inventory (STEP 8); the
decision-criteria axis is the checklist your argument must visibly tick.
(`references/audience-and-message.md` §3.)

### STEP 3 — Defeat the Curse of Knowledge

Once you know something you cannot reconstruct not knowing it, so you default to undecodable
abstraction. The cure is **concreteness**: drop from claim → sensory, picturable example → then
generalize. If you cannot name a 90-second concrete instance of your central claim, the curse is
biting. (`references/audience-and-message.md` §4.)

### STEP 4 — The cognitive gap and the governing sentence

```
COGNITIVE GAP = goal state (what they must believe to act) − initial state (what they believe now)
```

The gap is an **ordered list of belief-shifts** — the outline before it is an outline. Compress the
whole talk into **ONE governing sentence**: the *answer to the audience's real question*, not a topic
label. Run it past "So what?" until it states an action-relevant claim. If you cannot write it crisply,
stop — do not proceed to slides. (`references/audience-and-message.md` §5–6.)

### STEP 5 — Order

Sequence the belief-shifts by the audience's **next question**, not your discovery order. Default to
**answer-first** (governing sentence up front, pyramid of support); flip to question-first only when
the conclusion would be rejected cold. (`references/structure-and-narrative.md`.)

### STEP 6 — Slides

Default deck = **assertion-evidence**: each slide title is a full-sentence claim, the body is visual
evidence for it, and reading the titles top-to-bottom reconstructs the whole argument (the titles-only
test). One message per slide. Cut to fit; never shrink the font.
(`references/slides-visuals-cognitive-load.md`.)

### STEP 7 — Per-context skeleton

Instantiate the skeleton for your context (research talk, investor pitch, exec readout, thesis defense,
demo, keynote) and apply the default that **flips** there. Don't improvise the structure.
(`references/context-playbooks.md`.)

### STEP 8 — Open / close / Q&A

The open buys the right to be heard; the close fires the one decision (the explicit ask — never "Thank
you, any questions?"); Q&A is where the decision most often actually moves. Build the objection
inventory first; route each doubt to *pre-empt in the main line* or *hold for a backup slide*.
(`references/delivery-objections-and-evidence.md`.)

### STEP 9 — Diagnose the binding constraint, then compress

Persuasion fails through ONE binding constraint at a time — **ethos** (credibility), **pathos**
(feeling), or **logos** (argument). Diagnose which is the bottleneck and spend your scarce attention
there, not uniformly. Then treat the time cap as a compression discipline: cut, rehearse out loud
against the clock, budget by leverage. (`references/delivery-objections-and-evidence.md`.)

## The non-negotiables (apply at every STEP)

- **Backward from the audience.** Order by *their* questions and starting state, never the speaker's
  discovery order or section list.
- **One goal, one governing sentence.** If you can't name the single state-change and the single
  sentence, you don't have a talk yet.
- **Concreteness over abstraction**, applied *repeatedly* (claim ↔ example, what-is ↔ what-could-be).
- **Pre-state limitations; never let them be discovered.** A limitation you raise builds ethos; one
  extracted under questioning is a wound. Contradiction-laundering always surfaces in Q&A.
- **Default + escape hatch, not a menu.** Each rule below has a regime where it flips; know both.
- **Never repeat a presentation myth as true** (22× stories, 7-38-55, Dale's Cone, the "18-minute
  attention limit," oxytocin-as-persuasion). They are debunked in
  `references/delivery-objections-and-evidence.md`; cite the accurate version or drop the number. Tool
  names and contested empirical specifics live in references with citations — never as undated
  assertions here.

## Execution model — design solo, audit fan-out

Operating guidance from a frontier model (Fable 5, 2026-07) to whatever model runs this skill: a talk
is **ONE argument for ONE real room** — STEPs 0–5, 7, 9 and the slide DRAFT are one-context by
construction; a harness buys the *read-only audit layer* and objection red-teaming, never the design.

| Mode | What | Bar |
|---|---|---|
| **SOLO — never shard** | medium gate · goal · gap · governing sentence · ordering (vertical logic is a whole-tree test) · slide drafting (the titles-only test is a whole-deck property) · binding-constraint diagnosis | this file |
| **FAN-OUT, read-only — only after the front-half pre-flight passes** | per-slide checklist audit; deck-wide myth/attribution grep with primary-source fetches; lens-diverse persona red-team that *proposes* doubts-inventory candidates, confirmed against the user's room model | slide-layer pre-flight + anti-pattern table (`references/slides-visuals-cognitive-load.md`); myths-vs-evidence ledger (`references/delivery-objections-and-evidence.md`), refutation-prompted |
| **PIPELINE** | one backup slide per routed objection | `references/delivery-objections-and-evidence.md` |
| **NOT delegable** | the audience's initial state — elicit the five axes from the user; agents cannot know this room — and rehearsal / the real room | — |

Anti-patterns: **simulated-audience laundering** — persona-agent reactions ("N agents found it
convincing") treated as facts about the room; they are candidates only. **Sharded spine** —
governing sentence / staircase / titles merged from per-section agents; the two global tests exist
to catch exactly the shard-boundary defect. **Folklore-confirming verifier** — an agent asked to
SUPPORT a choice returns training-data folklore; verifiers are refutation-prompted against the ledger.

Scale: single talk → fully solo, run the checklists yourself; large deck / defense backup arsenal →
solo design + read-only audit fan-out + per-objection pipeline; multi-deck portfolio → audit fan-out
per deck, design still solo per deck. No harness → same order, the audits become separate focused
passes. Harness mechanics → declared local configuration and executable checks; wording-level prose audits → linting-prose.

## Pointers (depth lives in references/)

- **`references/audience-and-message.md`** — STEPs 0–4: the one goal, the five initial-state axes, the
  Curse of Knowledge, the cognitive gap, the governing sentence, the doubts inventory, and three full
  worked examples (research talk, pitch, exec readout).
- **`references/structure-and-narrative.md`** — STEP 5: ordering machinery (Minto pyramid, SCQA/SCR,
  BLUF, strategic narrative), the answer-first default and when it flips, with the attribution/dating
  caveats.
- **`references/slides-visuals-cognitive-load.md`** — STEP 6: assertion-evidence, cognitive load,
  Mayer's multimedia principles, the redundancy effect and its reversals, figure-as-compression, and
  matching slide density to expertise.
- **`references/context-playbooks.md`** — STEPs 0 & 7: the medium decision and per-context skeletons
  (research talk, investor pitch, exec readout, thesis defense, demo, keynote) with the rule that flips
  in each.
- **`references/delivery-objections-and-evidence.md`** — STEPs 8–9: open/close/Q&A as designed
  structure, the ethos/pathos/logos binding-constraint diagnosis, compression, and the consolidated
  **myths-vs-evidence ledger** (every contested number and debunked claim, with citations).