The worksheet, the prompts, and the example the guide points to
This companion holds the working material for the guide: a one-page worksheet for sorting what an AI conversation is about, six starter prompts as plain text you can edit, and a small invented example with a script that checks it. Everything in the example is fictional, and nothing in it needs a graph database, a hosted AI service, or special software. The worksheet and prompts are on these pages. The dataset and script are in the kit, which does not yet have a public address.
Start Here
-
Open the classification worksheet and run one real subject through it.
-
Copy Starter Prompt 1 from the starter prompts, put your notes in place of the bracketed text, and paste it into any AI assistant.
-
Read the four lists it returns, and answer its questions.
The example
The example is a fictional clinic, Cedar Valley, that buys a referral service from a fictional vendor. It holds a case fact sheet, a dataset that records the clinic, its people, the service and three ideas as Eidolons, Ergons, and Noemas, a set of small rules a record must meet, and a script that checks them. The dataset defines no term of its own. It uses the Intellectual Frontiers ontology, which is published in the public root repository, and it does not copy it.
To run the checks you need Python 3.11 or later and a tool that runs a script with its dependencies, such as uv. From the kit folder, run uv run --with rdflib --with pyshacl scripts/validate.py. The script writes nothing. It checks syntax, shape, labels, relationship kinds, and agreement with the fact sheet on figures, dates, and states. It does not read the meaning of a description. A person checks whether a statement is true.
What is automated and what a person does
| Automated | A person |
|---|---|
Every Turtle file parses |
Decides what each subject is |
Records meet their shapes, and the failing examples fail |
Checks that each fact is true and current |
Each reflection matches its subject’s type, in the script’s small table |
Approves classification choices that are open |
Counts, figures, dates, states, and sources agree with the fact sheet |
Reads every AI reply before saving it |
The related guide
Pocket Guide 001, How to Describe Complex Ideas to Your AI, shows how to give records like these to a builder as a specification and a model of the world, and how to keep both current.