WFGY/Avatar/docs/quickstart.md

<!--
AI_NOTE_START

Document role:
This page is the fastest practical entry into WFGY 5.0 Avatar.

What this page is for:
1. Give the shortest path from first download to first useful output.
2. Help new users try the two official Atlas starter builds.
3. Explain how to evaluate the first run without overthinking the system.
4. Show the first tuning loop through WFGY_BRAIN.

What this page is not:
1. Not the full architecture page.
2. Not the full research page.
3. Not the full multilingual report.
4. Not a complete tuning manual.
5. Not a proof page for all current and future capabilities.

How to use this page:
1. Pick one Atlas build.
2. Paste it into a strong AI system.
3. Run one real task.
4. Compare the feel of the output, not just the wording.
5. Tune WFGY_BRAIN if needed.
6. Save your edited version and keep iterating.

Important boundary:
This page is designed for first use.
It is not a claim that every language, every mode, or every future avatar direction is already fully finalized.

AI_NOTE_END
-->

# ⚡ Quickstart

This is the shortest useful path into **WFGY 5.0 Avatar**.

Do not start by reading every research page.  
Do not start by over-optimizing settings.  
Do not start by trying to understand the whole system in theory.

Start by running one real task and feeling the difference.

---

## 🧭 What You Need

You only need three things:

1. one Atlas build
2. one strong AI system
3. one real prompt you actually care about

The goal of the first run is simple:

**see whether the language behavior feels more alive, more governed, and more reusable**

---

## 🫂 Pick an Atlas Build

WFGY 5.0 Avatar currently starts with two official Atlas builds.

| Build | Best first use |
|---|---|
| **Atlas · PSBigBig** | You want a steady, clear, rational-friendly build for writing, analysis, and public-facing tasks |
| **Atlas · MiniPS** | You want a warmer, more companion-forward build for conversation, reflection, and support-style interaction |

Target download locations:

- `Avatar/downloads/PSBigBig.txt`
- `Avatar/downloads/MiniPS.txt`

If you are not sure which one to try first, start with **Atlas · PSBigBig**.

---

## 🚀 The 60-Second Flow

### Step 1. Open one build file

Open either:

- `PSBigBig.txt`
- `MiniPS.txt`

Read only enough to confirm which build you chose.  
Do not try to memorize everything.

### Step 2. Paste it into your AI setup

Paste the selected TXT into the system-level or equivalent high-priority instruction area of your AI tool.

Use the cleanest setup available to you.  
For the first run, avoid stacking too many extra prompts on top of it.

### Step 3. Give it one real task

Use one prompt that actually matters to you.

Good first tasks:
- write a public post
- rewrite a paragraph to sound more alive
- explain a hard idea in a more human way
- answer like a calm, reliable assistant
- turn notes into a stronger public statement

Bad first tasks:
- vague tests with no real stakes
- random roleplay
- asking it to do everything at once
- comparing ten unrelated tasks in one run

### Step 4. Judge the first run correctly

Do **not** only ask:

- is it smoother
- is it prettier
- is it more impressive

Ask instead:

- does it feel more present
- does it feel less generic
- does it surface the real point faster
- does it sound more like a governed voice than a polished AI blur
- does it feel reusable, not just lucky

### Step 5. Tune `WFGY_BRAIN` if needed

If the output is too polished, too cold, too soft, too abstract, or too generic, do not rewrite the whole system.

Instead, tune the brain layer.

Ask the AI to help you rewrite the `WFGY_BRAIN` block in the direction you want.

Examples:
- make this build less polished and more grounded
- keep it clear, but reduce quoteable polish
- make it warmer without becoming sugary
- increase public-writing force without making it slogan-heavy
- keep the rational tone, but make it feel less distant
- keep the companion feel, but reduce softness and increase backbone

### Step 6. Save the edited version

When you get a version you like, save it as a new local variant.

That is the real beginning of Avatar use.

You are no longer just trying a prompt.  
You are starting to build a reusable language runtime.

---

## 🧠 First Prompt Suggestions

If you want clean first tests, start with one of these.

### Public writing
> write a short public post about why most AI writing feels too polished and too empty

### Rewrite
> rewrite this paragraph so it sounds more alive, more grounded, and less like default AI output

### Analysis
> explain the core weakness of this idea in plain English without sounding sterile

### Companion-style interaction
> respond to someone who feels stuck and overwhelmed, but do not sound generic, sugary, or fake

### Rational-friendly interaction
> answer this problem calmly, clearly, and with structure, but do not sound robotic or distant

---

## 🔧 First Tuning Loop

The easiest way to feel the system is to run one task twice.

### Round 1
Use the original Atlas build.

### Round 2
Use the same build, but ask the AI to rewrite `WFGY_BRAIN` in one direction only.

Good one-direction edits:
- less polished
- more grounded
- stronger public force
- warmer but not softer
- calmer but not flatter
- more companion-like
- more rational-friendly
- less quoteable
- less over-complete

Then run the exact same task again.

This comparison teaches you more than reading ten theory sections.

---

## 👀 What To Look For

You are not trying to see whether the system sounds fancy.

You are trying to notice whether it changes the behavior surface in a meaningful way.

Good early signs:
- the opening reaches the point faster
- the tone feels more alive
- the text feels less generic
- the output is easier to reuse
- the build keeps its identity across more than one task
- the writing feels less like “AI polishing” and more like “governed behavior”

Bad early signs:
- everything becomes prettier but emptier
- the build only works on one lucky task
- the personality depends on obvious gimmicks
- the writing becomes noisy instead of alive
- the output feels over-shaped
- the system sounds like a polished host, not a usable avatar

---

## 🌍 First Multilingual Trial

Do not begin with a 10-language stress test.

For your first multilingual trial:

1. choose one short prompt
2. run it in one language you know well
3. run it again in one second language
4. compare behavior, not just literal wording
5. check whether the voice survives the shift

Good starter languages for public demos:
- English
- Traditional Chinese
- Japanese
- French
- Thai

Remember the current public direction:

**multilingual Avatar is not simple translation**

It is about behavior shaping, calibration, and controlled drift.

---

## 🧪 What This Quickstart Does Not Prove

A good first run does **not** prove:
- full multilingual closure
- full production readiness
- perfect style reconstruction
- universal task stability
- completed future avatar infrastructure

This page is about first contact, not final proof.

That honesty matters.

---

## 📚 Where To Go Next

### If you want to tune behavior
Go to [🛠️ How to Use WFGY_BRAIN](./how-to-use-wfgy-brain.md)

### If you want examples first
Go to [🫂 Demo Hub](../demos/README.md)

### If you want the current credibility surface
Go to [📊 Eval Hub](../eval/README.md)

### If you want the deeper system story
Go to [🧪 Research Hub](../research/README.md)

---

## 🔗 Quick Links

- [🏠 Avatar Home](../README.md)
- [🫂 Demo Hub](../demos/README.md)
- [🛠️ How to Use WFGY_BRAIN](./how-to-use-wfgy-brain.md)
- [📊 Eval Hub](../eval/README.md)
- [🧪 Research Hub](../research/README.md)
- [⬆️ Back to WFGY Root](../../README.md)