CWN Agent Execution Boundary WorkKit, version 2
===============================================

This is the free kit from the Cedar House episode with Nigel and Vero. Use it on your own or with the owner of the service.

What is inside
--------------
  CWN_Agent_Execution_Boundary_Worksheet_SCREEN.pdf: the 4 page worksheet, dark version for your screen.
  CWN_Agent_Execution_Boundary_Worksheet_PRINT.pdf: the same worksheet, light version for your printer.
  prompts/PROMPT_1_BOUNDARY_REVIEW.txt: the review prompt to paste into your assistant.
  prompts/PROMPT_2_EVIDENCE_REVIEW.txt: the follow up prompt that writes a 1 page evidence review.
  AGENT_EXECUTION_BOUNDARY_WORKSHEET.md: the 10 questions as plain text.
  EVIDENCE_WORKSHEET.csv: the 12 field evidence map, with columns for how you know and for the state.
  VERO_CANCELLATION_CHECK_KIT.md: the 7 review steps and the 4 states.
  field-kit/cancellation_check.py: a small local test. No network and no AI model.
  checks/test_cancellation_check.py: 21 checks that show how the test behaves.

How to start
------------
  1. Pick 1 consequential action in your own workflow. Write the 1 decision this review should change, and by when.
  2. Answer the 10 questions. Write UNKNOWN when you cannot show an answer.
  3. Fill 1 row of the evidence map for that action. Mark how you know each answer.
  4. Paste prompt 1 into your assistant with documents that describe the workflow. Use invented or cleaned records. Do not paste personal information.
  5. Paste prompt 2 to turn the table into a 1 page evidence review.
  6. Run the local test. You need Python on your computer. The command is: python field-kit/cancellation_check.py
  7. Take the evidence review to the people who own the records.

The 4 states
------------
  - Found in the system's own record.
  - Looked for and not there.
  - Could not be checked.
  - Nobody has checked.
  - An unknown is not a pass and it does not count as 0. It is the next thing to test.

What this is and is not
-----------------------
  - This is a teaching aid. It does not give a safety rating, a legal opinion or an insurance view.
  - The test is a toy that runs on invented data. Do not connect it to a live system.
  - The 2 independent sources and the 300 second window are teaching numbers. They are not advice for a real emergency system.
  - The test shows why 3 copies of 1 report are not 3 sources. It does not measure how often a model makes a bad decision.
  - Your assistant judges what you paste. It cannot check your systems.
  - Cedar House is a fictional scenario with AI-generated hosts. Mercer and Rosa are fiction.

Where the approach comes from
-----------------------------
  The Agent Execution Boundary approach is described by Cyber Warrior Network at https://cyberwarriornetwork.com/agent-execution-boundary-review. It is free to read: the definitions, the 4 states and answers to common questions.

Free, for your own design reviews.
