How the ISATVON Framework Works

The process

From Raw Request to Reviewable Response

Start With a Raw Prompt

Write your task in normal language. It can be short, rough or incomplete.

→

Structure It With ISATVON

Organise the request into seven sections covering instructions, sources, execution methods, tools, variables, outcomes and reporting requirements.

→

Use Your Preferred AI Tool

Paste the structured prompt into ChatGPT, Claude, Gemini, Perplexity, Copilot, Grok or another capable AI platform.

→

Review a Transparent Response

The AI response should explain what it understood, which sources it used, how it verified the work and what assumptions it made.

The framework

The Seven Sections

ISATVON stands for Instructions, Source, Automation, Tech Stack, Variables, Outcome and Notification. Each section works in two directions: it tells the AI how to perform the task, and it tells the AI what it must report in the response.

Instructions

In the prompt: Role, task and non-negotiable instructions.
In the response: The task as the AI understood it.

Source

In the prompt: Approved sources and information the AI must not invent.
In the response: The sources actually used.

Automation

In the prompt: Workflow steps, checks and self-verification requirements.
In the response: How the result was produced and verified.

Tech Stack

In the prompt: Allowed, required and prohibited tools.
In the response: The tool policy followed during execution.

Variables

In the prompt: Length, tone, audience, format, language, depth and fallback rules.
In the response: Constraints followed, missed or modified, with reasons.

Outcome

In the prompt: The required output structure and acceptance criteria.
In the response: The completed deliverable.

Notification

In the prompt: Assumption, confidence and omission reporting requirements.
In the response: Assumptions, confidence levels, missing inputs and limitations.

Reading an ISATVON Response

Every ISATVON prompt’s O section asks for the reply in the same structure, so the response is auditable at a glance.

isatvon-response.md

IThe model solved the wrong problem: visible in one sentence, before you read the rest.

SAnswers grounded in training data instead of your supplied context announce themselves.

A“Verified: nothing” means unchecked claims are visible.

VSilently broken constraints (length, tone, scope) are reported.

OThe deliverable, cleanly separable from the meta and easy to copy out.

NHidden assumptions and low-confidence guesses are disclosed, not dressed up as facts.

Check I first: a wrong restatement means a wrong answer; stop and re-prompt. Scan V and N for broken constraints and assumptions you disagree with. Only then read O.