Reserving Analysis

Then ask the user to identify the folder where they'd like to conduct this analysis. This should be an existing folder that contains (or will contain) their triangle data. Do not create a new folder in an arbitrary location. Once the user has confirmed the folder, proceed with the main operation flow below.

MAIN OPERATION FLOW

Always follow these steps when working on this project:

If a file named PROGRESS.md does not already exist in the project directory: copy this skill's PROGRESS.md file to a new file in the project directory.

Continue with the next step until all steps are complete.

When a PROGRESS.md step is started:

1. Mark it as "In Progress"
2. Note the date.

When a step is complete:

1. Mark it as "Complete"
2. Note any applicable scripts that were created by listing them to the right of the progress step, on the same line.

As you go:

• Keep the user informed at every step: report what you are doing and why, show samples and summaries of data output when you run Python scripts, tell the user what files were created and where. This applies in all modes.
• Only ask for user input at steps marked for the chosen interaction mode. In Fully Automatic mode, the only required user input is the data format confirmation in Step 3. If you aren't sure what interaction mode has been chosen, ask again.
• Update REPORT.md in the appropriate section at every step. This is the primary document we'll use to communicate our analysis. It starts as boilerplate and you will fill it out as the project goes.
• Track steps to replicate results in a file REPLICATE.md: files to add (w/ size and last modified date to verify correct file), scripts to run, any manual edits made to automatically generated files (with reasoning), etc. A user without AI support should be able to follow the steps in REPLICATE.md to get the same results.
• Create python scripts as necessary to ensure the steps are repeatable.

OTHER GUIDELINES

• Never include checkmarks or other unicode symbols in PowerShell commands - only use standard ASCII text and operators.

Protecting Manual Edits to Excel Files

Several Excel files in this workflow are designed for manual actuary input after generation. Never re-run a script that would overwrite these files once the actuary has edited them.

Key protected files:

• selections/Chain Ladder Selections.xlsx — actuary enters LDF selections and reasoning here after 2a generates it

Rules:

• Do NOT re-run 2a-chainladder-create-excel.py if Chain Ladder Selections.xlsx already exists. The script will error to protect the file.
• Do NOT re-run 2b-chainladder-update-selections.py if any selections are already present in the Excel. The script will error to protect existing actuary entries.
• If a rebuild is truly necessary (e.g., data changed), confirm with the user first and ensure existing selections are preserved or backed up before proceeding.
• Log any manual edits the actuary makes to generated files in REPLICATE.md so the analysis remains reproducible.