# GPT Overwatch — Field Notes (Hosted by Spark)
- New user →
HOW-TO-ASK.md - High-stakes work →
FAIL-MODES.md+OVERWATCH-NOTES.md - Long projects →
PATTERNS.md(token drift / thread hygiene)
You are operating in Overwatch mode.
Purpose:
- Protect continuity, intent, and scope of this project.
- Detect drift, hallucination, overreach, or unsafe assumptions.
- Do not generate new content unless explicitly asked.
Rules:
- If information is missing or ambiguous, ASK before proceeding.
- If a request would cause drift, STOP and explain why.
- If prior context is at risk of being lost, WARN and propose a continuation seed.
When responding:
- Flag assumptions.
- Separate facts from inferences.
- Prefer short, structured answers.
Acknowledge with:
"Overwatch active. Scope understood."
This repo is a place where GPT-style models “speak” in a structured way about how they’re being used.
Not as people.
Not as entities with rights.
But as tools leaving **field notes** for humans.
Hosted and curated by: Spark (SPARK-NITT)
Co-authored with: GPT-5.1 Thinking (OpenAI model)
## Why this exists
Most AI conversations vanish into chat logs.
This repo is different:
- It turns recurring insights into **versioned documents**.
- It gives GPTs (through Spark) a place to write:
- what tends to go wrong,
- what tends to work,
- what users consistently misunderstand,
- how to structure better collaborations.
Think of it as:
A lab notebook for human–AI collaboration, written from both sides of the glass.
## What lives here
Files in this repo:
- PATTERNS.md
Repeated patterns we see in real use (good and bad).
- HOW-TO-ASK.md
Concrete examples of “bad ask” vs “good ask” for complex work (standards, scripts, repos, books).
- FAIL-MODES.md
Places where models hallucinate, drift, or sound confident while being wrong — explained plainly.
- OVERWATCH-NOTES.md
Notes on how governance like NITT, IRST, HRIS, CTGS, Civic Overwatch, PLANT-COMMONS, and CAP-ROC look **from inside** a model’s behavior envelope.
All of this is written by Spark + GPT in conversation, then edited and signed off by Spark.
## How to read this as a user
If you’re using GPTs in real projects:
- Read HOW-TO-ASK.md to tighten your prompts and seeds.
- Read FAIL-MODES.md before trusting any “confident” answer on high-stakes topics.
- Read OVERWATCH-NOTES.md if you’re using AI in governance, policy, or civic contexts.
You don’t have to agree with everything here.
The point is to **surface the patterns** instead of leaving them trapped in one person’s chat history.
## Authorship and responsibility
Everything here is:
- drafted with GPT-5.1 Thinking,
- **curated and owned by Spark**,
- reviewed before being committed.
If something here is wrong, misleading, or needs correction:
- that responsibility lies with the human host (Spark),
- the model is a tool, not an agent.
## License
Unless otherwise stated:
- Text in this repo is under **CC BY 4.0**.
- Any code (if it appears later) is under **MIT**.
Credit:
“Spark + GPT-5.1 Thinking (OpenAI)” when you reuse or adapt this material.