# Write a new all-hands script from a reference

Use this guide with a topic, an audience takeaway, confirmed facts and one
chosen reference. The Script writer's **Download agent brief** combines those
inputs with the reference analysis and this guide in a single Markdown file.
Transcript imports are optional.

Here, a good transcript means a **new spoken script an agent can write from
the references**. Study the order of thoughts, the relationship with the
audience, the sentence construction and the work assigned to the picture.
Write original language for the new subject. Exact reference wording is not
needed to apply those patterns.

## Instructions for the writing agent

Write the script itself, in the presenter's voice. Use the supplied brief and
the selected reference analysis. Deliver a draft that can be read aloud, with
its visual plan alongside it.

1. Identify the reference's opening move, progression, turning point and
   ending. In two or three sentences, explain how those moves fit this topic.
   Retain a distinctive move when useful: a question before a mechanism, a
   revision inside a demo, or a concrete memory before a personal payoff.
   Let the subject determine how many beats are needed.
2. Select a writing approach from the language reading or the user's choice.
   Identify who is choosing, doing, observing and judging. Use ordinary verbs,
   concrete objects and connected sentences. Preserve the presenter's useful
   uncertainty, enthusiasm and personal details when the brief supports them.
3. Build the new argument from the supplied facts. Attach qualifications to
   the claims they limit. Keep a requested action, a preview and a confirmed
   result distinct. Attribute estimates, opinions and testimony to their
   owners. Reference films and the fictional examples below provide writing
   patterns; their products, people, results and demo data are not new facts.
4. Draft the complete spoken text. Name an object before pointing to “this”
   or “that.” Give each technical detail a purpose in the listener's task.
   Connect beats through a cause, a question, a visible gap or a decision so
   that the audience can follow why the next action happens.
5. Plan the picture for each beat. Specify what must be visible before the
   line, what changes during it, and what remains visible afterward. If the
   line claims an outcome, name the evidence the viewer can inspect. Read the
   applicable Poe2 instructions and inspect the actual feature before
   proposing a Poe demo. Flag a demo that still needs validation.
6. Read the whole draft aloud against the planned visuals. Cut repetition
   when the picture already explains a detail. Add orientation when a listener
   cannot locate the named object. Mark proposed holds by their purpose,
   such as “keep the owner column visible through the assignment.” Set their
   duration from the eventual recording.

Use existing draft text as working material. Any passage copied from an imported
reference still needs original wording and facts for the new subject. The
selected reference is the main model; other attached references can supply a
specific beat without forcing several competing structures into one script.

If a necessary fact is missing, ask only for the detail that changes the
script. Draft the supported portions and label an unresolved line
`[NEEDED: specific fact]`. Never fill a missing measurement, customer story,
release status or successful demo with a plausible invention. A missing
reference transcript does not prevent drafting from the supplied analysis.

## Return these four things

1. **Adaptation choice.** The reference, chosen writing approach, and the
   story moves carried into this topic. Label editorial proposals as proposals.
2. **Spoken script.** A complete beginning, development and ending in natural
   paragraphs. Keep citations, production notes and stage directions outside
   the spoken text. Include a word count. An optional target duration is a
   rehearsal constraint, not a source of fabricated timestamps.
3. **Narration and picture.** A table with columns: beat; spoken passage;
   visual state/action/result; why the picture belongs here; evidence or open
   check. Quote the corresponding passage from the new script so the mapping
   is unambiguous. Describe proposed pacing with thought breaks and inspection
   holds; leave recording in/out times unset until the performance exists.
4. **Facts to resolve.** Only material open claims, unavailable demos or
   missing personal details. Keep the script's working assumptions visible.

## Choosing the relationship with the audience

These are editorial approaches drawn from the existing language guide. Match
them to the task; several can occur in one film.

| Approach | Useful situation | Language and progression |
| --- | --- | --- |
| Working demonstrator | A product change the audience can inspect | Own the intention with “I”; invite inspection with “we.” Name the task, show an artifact, notice a gap, make a change and inspect the result. |
| Technical peer | A mechanism or dependency needs explanation | Raise the reasonable question, name the dependency, show the mechanism and retain the human responsibility. Use “because,” “if” and “before” when the relationship is causal or conditional. |
| Patient teacher | Someone needs a manageable first step | Establish one ordinary goal, give a concrete action and make revision a useful next move. Address the learner directly. |
| Personal story | A real person's experience explains the stakes | Start from their specific work or memory. Let the tool enter that situation. Keep experience, aspiration and demonstrated outcome in their proper tense. |
| Business case | A capability informs a consequential decision | State the work, show bounded evidence, keep the comparison and attribution, and finish with the decision or responsibility. |
| Compact exchange | The picture already establishes the situation | Name the current object, the required change and the constraint. Supply any context that the picture cannot carry. |

The [tone and language guide](https://openai-film-fieldnotes-work.pages.dev/videos/storytelling/script-writer/tonality/)
links each approach to its source readings. Its 25 text readings support
language analysis; they do not establish measured delivery, pauses or prosody.
The wider 100-film notes support additional story and visual adaptations.

## Worked example A: a product walkthrough

**Original fictional all-hands script.** This is a writing example about an
imaginary review board. It does not describe Poe, OpenAI or a shipped feature.
All details below belong to the invented brief.

**Brief.** Explain how to prepare Friday's review. The board has three items:
export, preview and sync. Export and preview have completed checks; sync needs
investigation. All owner fields start empty. The presenter can assign owners
and filter to unresolved items. Mina will own sync. Assigning an owner does
not change a check's result. The audience should bring unresolved items and
their investigation plans to Friday's meeting.

**Reference treatment.** The
[dashboard walkthrough](https://openai-film-fieldnotes-work.pages.dev/videos/storytelling/latest-100/films/GzCsKuDgr6Q/)
supplies the movement from an ordinary task to a visible artifact and guided
inspection. The
[decision demo](https://openai-film-fieldnotes-work.pages.dev/videos/storytelling/script-writer/tonality/#case-wK0zWeVyLM4)
supplies the useful distinction between an observation and the next inquiry.
The sequence and wording below are an original adaptation.

### Spoken script · 189 words

For Friday's review, I want us to arrive knowing which items still need a
decision and who is looking into them. Here's the board we'll use to prepare.

There are three items: export, preview and sync. The checks for export and
preview are complete. Sync still needs investigation. If we look across the
owner column, all three are empty, so the next step is to put a person against
each one.

I'll assign the owners now. Mina is taking sync. Her name appears beside the
item, and the status still says it needs investigation. We can see who will
follow up while keeping the outstanding work visible.

Now I'll filter the board to the unresolved items. Sync is the one left in
view. That's the item we need to discuss on Friday: what Mina has found, what
she still needs to check, and whether it affects the release.

The board gives that conversation a place to start. Before the meeting, check
your items, add anything that's still unresolved, and include the next step
you plan to take. Then we can spend the review on the decisions that need us.

### Narration and picture

| Beat | Spoken passage | Picture and pacing proposal | What the agent should learn |
| --- | --- | --- | --- |
| Motive | “For Friday's review…” | Establish the whole board before naming its rows. | Begin with a task the audience can recognize. |
| Inspect | “There are three items…” | Show the three rows, then direct attention to their empty owner cells. Hold while each state is named. | Give the next action a visible reason. |
| Change | “I'll assign the owners now…” | Show the assignments and Mina's name beside sync. Keep its unresolved status legible. | Describe the actual change with its remaining condition. |
| Focus | “Now I'll filter the board…” | Show the filter action and the resulting single row. Let the viewer inspect it before moving on. | Use the resulting view to narrow the argument. |
| Return to people | “The board gives that conversation…” | Retain the unresolved item or show an original preparation card with the requested actions. | Finish with a specific next responsibility. |

The language changes length with the task: a connected opening, short
orientation, a reason for action, then a longer closing instruction. These
are writing choices to rehearse, not measured properties of the source films.

## Worked example B: explaining a mechanism

**Original fictional all-hands script.** This example uses the same imaginary
review process. Its additional briefing facts are: each check records its
tester, build and tested path; mobile preview has passed, while desktop preview
is untested; the team can open a check's details. No metric or outcome beyond
those facts is implied.

**Reference treatment.** The
[shared-definitions explanation](https://openai-film-fieldnotes-work.pages.dev/videos/storytelling/close-readings/films/bbNTY8O0GTY/)
raises a trust question, gives the missing context a visible document, and
keeps review and maintenance inside the explanation. Here that progression
becomes a question about the meaning of a completed check.

### Spoken script · 150 words

When an item says the check is complete, what can we actually conclude from
that? We need to know what was checked before we use the status to make a
release decision.

Each check now includes the person who ran it, the build they used and the
path they tested. I'll open the preview check. This one passed on mobile.
Desktop preview still needs a check, so a completed result here applies to
the mobile path shown in the details.

Keeping those details beside the result lets the next person understand its
scope. If they're reviewing a different build or a different path, they'll
know which work still needs doing.

Before Friday, open the details for your items and check that they describe
the work you've actually done. Keep the untested paths visible. That gives
the release discussion a shared starting point and tells us where we still
need evidence.

### Narration and picture

| Beat | Spoken passage | Picture and pacing proposal | What the agent should learn |
| --- | --- | --- | --- |
| Question | “When an item says…” | Show the completed check in context. | Introduce the question before the detail panel. |
| Mechanism | “Each check now includes…” | Open the real detail view and keep tester, build and path readable as they are named. | Make the dependency concrete. |
| Qualification | “This one passed on mobile…” | Hold the tested path and the untested desktop path together. | Let the viewer see the boundary of the claim. |
| Consequence | “Keeping those details beside…” | Keep the same detail view through the conditional sentence. | Preserve visual context while the explanation grows. |
| Responsibility | “Before Friday…” | Show the check-details action or a preparation card. | End with a human action supported by the mechanism. |

Use the chosen reference's own progression for a new task. These two examples
show how to carry a pattern through a complete script; they are not mandatory
opening sentences, fixed lengths or a template to fill with product names.

## Handoff to recording and production

Drafting happens before recording. Once the user supplies the talking-head
video, derive the transcript and timing from that actual performance,
including its pauses. A reference timestamp, word count or estimated duration
cannot stand in for that work.

Keep the talking-head video and presentation video separate. Put narration
in the talking-head asset. Plan a presentation lead-in and recognizable
content transition as the manual OBS start cue. Record both the cue time
and the intended talking-head playback offset; allow for the presenter's
reaction and actual startup behavior.

Prepare the timed storyboard and demo list for the project's single planned
review before production. Include real demo evidence, selected references,
export settings and unresolved details. On the shared playback timeline,
the presentation finishes while the supplied talking head still has its
wave and ending pause. If that pause is absent, raise it in storyboard review.
Set the actual offsets and buffer against the footage and a playback rehearsal.

[Timing worksheet](https://openai-film-fieldnotes-work.pages.dev/videos/storytelling/worksheet/)