Delivery and Stage Operations
🗒️ Description
Everything that happens between arriving at the venue and leaving it, researched as the second wave behind Conference Talk Craft. Eleven techniques: how I sound and stand, what to do with nerves and with a failure, how a recorded talk differs from a room talk, and the operational items — stage tech, the question channel, the contract with the host, a resettable demo, a room check and a backup kit. The operational half is mostly practitioner consensus built from failure reports; the delivery half has more research behind it and, as usual, more folklore on top.
Siblings: Talk Narrative and Structure, Slides for Technical Talks, Practising and Reviewing Talks.
🧩 The techniques
Voice, body and nerves
| Technique | What it is | Priority |
|---|---|---|
| Voice | Four separately trainable controls. Pace: measured in syllables per second, not a borrowed words-per-minute figure. Pauses: 15–20 planned silent ones of 1–3 seconds around the key sentences. Projection: sized to the microphone, not the room. Fillers: cut down, not out, by habit reversal. Measure each from a recording. | Pauses high and nearly free; fillers medium |
| Body, stage position, pointer | Stand beside the screen where I can see both the slide and the room (Weatherman), not behind the lectern (Bunker). Let hands gesture. Point with the body or with a highlight built into the slide, and treat the laser as an occasional tool (Lightsaber), not a crutch (Laser Weapons). | Medium; slide-built highlights high for recorded talks |
| Speaking anxiety | In the last minutes, relabel the racing heart as readiness (“I am excited”) instead of trying to calm down, remind myself the audience sees far less than I feel, use slow long exhales if I want a physical lever, skip power posing. The one intervention with a large effect is repeated exposure. | Medium; cheap enough to always do |
On camera and when something breaks
| Technique | What it is | Priority |
|---|---|---|
| Camera versus room | Give a recorded conference talk to the room, not to the lens — serving a live and a remote audience at once is the Dual-Headed Monster. Then patch the six things the recording silently loses with cheap verbal habits: unrepeated questions, show-of-hands results, laser and hand pointing, room laughter, references to the day, and me outside the frame. Anything aimed mainly at online viewers is a separate recording. | High for recorded and reused talks |
| Recovery protocol | Written in advance: name the problem in one plain sentence, act (fix inside a 30–60 second time-box or switch straight to the fallback), bridge back with a prepared line, and never mention it again. | High |
| Stage tech | Settle five things before the day: which microphone (a headset or lavalier frees both hands), which clicker and which keys it sends, whether a confidence monitor exists and what it shows, the display’s native resolution and aspect ratio, and where the presenter view will physically live — my laptop in Extend mode, a monitor at the stage edge, or nowhere, in which case the notes go on a card. | High |
Question handling infrastructure
| Technique | What it is | Priority |
|---|---|---|
| Anonymous question channel | Let the room type questions on their phones during the talk, anonymously with upvotes (Slido, Mentimeter or the event app). A moderator reads the top ones at Q&A. The queue exists before the floor opens, nobody has to stand up, and I leave with a written list of every question. Needs the organiser’s agreement. | High for large, late or recorded talks where the organiser already runs a tool; otherwise medium |
| Contract with the moderator | One card agreed with the host in two minutes: whose clock counts and when my time ends by the wall clock; signals at 5, 2 and 0 minutes and where the host sits so I can see them; who picks questions and by what rule; who cuts a monologue; and 30 seconds for my close after the last answer. | High: the cheapest structural fix for timing and Q&A |
Demo and kit
| Technique | What it is | Priority |
|---|---|---|
| One-command demo reset | Every demo state reachable with one short command: a reset script that returns the repo, data and terminal to the starting state and checks that it did, a tagged checkpoint per demo stage, and long commands and prompts prepared in advance so I never type them live. | High for a live or hybrid demo that changes state |
| Room recon and tech check | Get into the actual room before the audience does, or into the changeover minutes, and run one fixed script: deck and video on the real screen at its real resolution, the microphone at stage volume, my smallest text read from the back row, where I stand and where my eyes go for notes and the clock, and whether the stage lights let me see the room at all. | High |
| Backup equipment checklist | A written list ranked by how badly each failure hurts: my own video cable and an adapter for the port the laptop really has, the charger, a clicker I know, and the deck in at least three independent places off the laptop — two USB sticks carried separately, a cloud link that needs no login, and a copy already with the organiser. Every copy holds the whole build. | High |
⚖️ What the evidence actually supports
Sound practice, wrong justification
| Practice | The usual reason | What the research says |
|---|---|---|
| Cut filler words | ”Fillers make people stop understanding you” and “they are just nerves” | They signal planning difficulty, and a filled pause is rated better than an equally long dead silence (Christenfeld 1995). The goal is fewer, not zero, and the replacement is a short intentional pause, not faster speech. Words after a silent pause are recognised better (MacGregor et al. 2010). The claim that fillers help recall is contested after a partial replication failed. |
| Speak at 130–150 wpm | An English coaching convention | No transfer to Polish, and words per minute mixes speaking time with pauses. Fluent delivery raises ratings, not learning (Carpenter et al. 2013). |
| Body language | ”55% of the message” (Mehrabian 7-38-55) | That study concerned contradictory emotional signals in single words, not informational talks. The practices — open body, free hands, deliberate position — are sound for other reasons: gesturing lightens the speaker’s working-memory load (Goldin-Meadow 2001) and an instructor pointing improves learning from video (Pi 2017). |
| Power pose backstage | ”It makes you confident” | Debunked as a performance or hormone intervention; later studies find only self-reported effects. |
| Reappraise arousal | ”Say ‘I am excited’ and you will perform better” | Brooks 2014 is real, but a 2025 direct replication in public speaking reproduced only the self-reported effects and observers rated the speeches no differently. Across reappraisal trials the performance effect is small (d = 0.23, Bosshard & Gomez 2024). Worth doing as a cheap mood intervention, not as a performance guarantee. Repeated exposure is the intervention with the large effect (Ebrahimi 2019, g = 0.74, rising to 1.11 at follow-up); virtual-reality exposure roughly matches real exposure. |
| Brief acknowledgement of a mishap | ”A mistake makes you more likeable” | The pratfall effect holds only for people already seen as competent. What is solid: observers judge blunders less harshly than speakers expect (Savitsky, Epley & Gilovich 2001), and only a repeated apology turns a glitch into the story of the talk. |
| Anonymous question tool | ”Upvotes surface the best questions” and “it fixes the gender gap” | One early upvote raised final ratings by 25% in a randomised study (Muchnik, Aral & Taylor 2013), so vote ranking snowballs; the moderator should not treat the top as fixed. Questions typed into a named chat showed the same gender gap as spoken ones (Zhang et al. 2023), so default anonymity is what matters. The classic disaster was a named Twitter feed projected behind the speaker without moderation: the failure was the projection, not anonymity. |
| Time signals | ”A time card keeps a speaker on time” | Enforcement does. How strictly limits were enforced was among the main predictors of overrunning across 11 conferences (Edlund et al. 2018). |
| Demo reset | ”It makes the demo deterministic” | False for demos built on LLM calls: a reset restores an agent’s inputs, not its outputs. |
Where the evidence is thin
- The anonymous channel has field reports from classrooms and one radiology conference (324 anonymous texted questions against 42 paper notes), small studies and vendor data, but no controlled comparison with microphone Q&A at a conference. Free tiers limit participants and moderation, and the limits change — check the plan before relying on one.
- The contract with the moderator rests on chair instructions from many conferences plus partial field evidence. One conference that offered the first question after every talk to a trainee saw women’s share of questions rise and stay higher (Telis et al. 2019). An astronomy-meeting result linking the chair’s gender to who asked did not replicate clearly elsewhere (Davenport 2014).
- Operations items — recon, backup kit, stage tech, demo reset — are practitioner consensus. The numbers attached (arrival times, minimum font sizes) are conventions, not findings. The back-row geometry is the exception: code and terminals need the farthest viewer within about four times the image height, large slide text about six times, from an AV standard.
- A demo reset has one contested point: one practitioner found a commit per demo step unworkable
because of constant rebasing and moved to numbered step directories. Also,
git clean -fdxdeletes caches I pre-downloaded on purpose.
Small facts that bite
- Presenter view needs Extend mode; venues default to Duplicate. HDMI audio routing is a separate setting. Narrating a silent demo clip removes one thing that can fail.
- A laser dot is invisible on LCD/LED walls and in slide-feed recordings. Build highlights into separate slides.
- A USB-C port carries video only with DisplayPort Alt Mode or Thunderbolt.
- On some clickers the start button sends F5, which reloads a browser deck and drops full screen; the Marp HTML deck has no handler for the blank-screen key.
- An HTML deck copied without its media folder plays black rectangles. A copy only counts if it is independent and complete.
- Standard Whisper drops filler words, so automatic filler counts look better than the truth.
- Advice about confidence monitors conflicts: one source says cover it, another says put one at the speaker’s feet. Decide per room.
🔧 Cheapest first moves
- Failure card: write the name-it line, the time-box and the bridge line for the three failures I fear most, then have someone take the clicker away mid-run and see whether the card is enough.
- Host card: the seven lines above on a single A5 card, handed over in person.
- Adapter drill: from the laptop in its bag to the first slide full screen on an external display with presenter view on the laptop, timed.
- Pause pass: mark 15–20 planned pauses in the presenter notes for one block and rehearse only that block.
📖 Further reading
- Conference Talk Craft — wave one: demo strategy, environment hygiene, data leaks, Q&A protocol
- Public Speaking — hub note
- Talk Narrative and Structure · Slides for Technical Talks · Practising and Reviewing Talks
- The Art of Public Speaking — Carnegie on voice and emphasis
- TED Talks — Anderson on stage fright and delivery
Template: knowledge_note_info