Fountain Coach Gitowner-controlled · read only

teatro-codex-macos-prompt-field-guide.git · decoding.md

teatro-codex-macos-prompt-field-guide.git / decoding.md

revision 0d2da6fea0d65a5f8a11dda39f0c834db1433ac7 · complete file

# Decoding

Here’s a **plain‑language, human‑first** way to use Teatro with a **director + actor + LLM** that’s about **decoding and shared insight**, not syntax.

---

## The point

Each short pass should end with the group saying the same simple things:

1. **What changed?**
2. **Why now?**
3. **Who holds power at the end of the beat?**
4. **What will we try next?**
5. **What single line or moment proves it?** (the “evidence”)

The LLM’s job is to **help you see** (observe neutrally), **offer options** (without insisting), and **remember** (capture decisions). Not to direct.

---

## The room (physical or digital)

* A shared board with **three columns**:

  * **What we saw** (facts only)
  * **What it might mean** (possible readings)
  * **What we’ll try** (one choice for the next pass)
* A **timer** (2–6 minute cycles).
* A simple **stage view** (if you have it) for light/tempo/camera tweaks—but treat them like **knobs**, not code.
* The LLM sits in one of **five hats** (you choose per pass):

  1. **Scribe** – captures facts and quotes exactly.
  2. **Mirror** – reflects patterns it noticed, in 3 bullets max.
  3. **Audience** – says how it felt as an ideal audience member.
  4. **Devil’s Advocate** – offers one strong alternative reading.
  5. **Archivist** – writes the one‑page **Beat Insight Card** at the end.

---

## One beat = one loop (5–7 minutes)

**1) Run (2–3 min)**
Director gives a plain note (“lighter start, heavier landing”). Actor plays. LLM is silent.

**2) Decode (2 min)**

* **Director:** “Let’s capture facts only.”
* **LLM (Scribe):** Bullet list of **observations** (no opinion):

  * “He kept eye contact until ‘…wrong.’”
  * “Voice dropped on the last clause.”
  * “She didn’t move; he closed distance by ~1 step.”
* **Actor:** Adds internal facts (“I felt stuck until the last line.”)

**3) Make meaning (1–2 min)**

* **LLM (Mirror):** “Two possible readings in 3 bullets total.”
* **Director:** Chooses **one reading** to test next.
* **Actor:** Names a **playable action** (“tease → confess”).

**4) Decide next try (30–60 sec)**

* Write **one** experiment in “What we’ll try.”
* Optional: tweak **one knob** (light/tempo/camera) to support the choice.
* **LLM (Scribe):** Saves the choice and the **evidence line/moment** you’ll watch for.

Repeat the loop once or twice; stop when you can answer the five questions cleanly.

---

## The Beat Insight Card (what you keep)

*(Written by the LLM as Archivist; one small page, plain language.)*

* **Beat title:**
* **We think this beat is about:**
* **Objective / Wall / Way / Risk** *(non‑jargon: Want / What’s in the way / How they try / What’s at stake)*
* **Evidence line/moment:** *(quote + what we saw)*
* **Power shifts:** *(before → after)*
* **Our choice for next time:** *(one sentence)*

This is the only “artifact” that matters. No DSL. No markup. Just the group’s shared understanding.

---

## What the LLM actually says (short scripts you can use)

**After a run (Scribe):**

* “Facts I saw (no opinions):

  1. He laughed before ‘…charm my way…’
  2. ‘…but I was wrong’ was quieter than the setup.
  3. She didn’t react until the last three words.”

**When invited to make meaning (Mirror):**

* “Two reads to consider:
  • The laugh is a shield that finally cracks on ‘wrong.’
  • He’s trying status play; the last words surrender it.”

**As Audience (emotion check):**

* “I felt amused → uneasy → tender. The turn landed on ‘wrong.’”

**Devil’s Advocate (one pushback):**

* “Counter‑read: There was no real surrender—volume dropped, but eye contact stayed dominant.”

**Archivist (end of beat):**

* “Beat Insight: He tries to keep status by joking, then surrenders it at ‘wrong.’ Evidence: the laugh → voice drop, plus closing the gap by one step. Power shifts to her.”

---

## Five plain prompts you can speak to the LLM

* “Capture **only what we saw**, no opinions, three bullets.”
* “Give us **two readings** that fit those facts—keep it short.”
* “As **audience**, how did that feel in one sentence?”
* “Offer **one** strong alternative that would change the ending power.”
* “Write the **Beat Insight Card** in plain language.”

---

## 20‑minute mini‑plan (no tech friction)

1. **Warm‑up (2 min):** Director states the beat in one sentence. Actor states *want* in five words or less.
2. **Pass #1 (3 min)** run; LLM scribe facts.
3. **Decode (3 min)** choose a reading; pick one action to test.
4. **Pass #2 (3 min)** run the choice; LLM audience reaction + mirror.
5. **Decide (3 min)** pick the keeper; lock the evidence line; capture power shift.
6. **Card (6 min)** LLM writes the Beat Insight Card; group edits one minute, done.

You can do three beats in an hour with this rhythm.

---

## Using Teatro’s knobs without making it “tech”

Keep the four core knobs as **dials**, not code:

* **Light** (warmer/cooler, brighter/dimmer)
* **Tempo** (slower/faster)
* **Camera** (closer/further, shared vs. solo)
* **Montage** (cut quickly / let it breathe)

Ask the LLM for **one** suggestion per knob in plain words only when you need it. Example: “Name one camera change that would underline surrender.” If you try it and it helps the reading, keep it; if not, drop it.

---

## What to avoid

* Long LLM essays. (Cap it: max 3 bullets, one sentence reactions.)
* Jargon. (Use **Want / Wall / Way / Risk**, not theory words.)
* Tool‑driven sessions. (Humans decide; LLM observes, offers, remembers.)
* Solving everything at once. (One insight per pass.)

---

### Bottom line

**Decoding** = short cycles of **seeing → meaning → choice → try → evidence**.
**Shared insight** = everyone can say the same one‑sentence beat, name the evidence line, and agree who holds power at the end.