ai-ledger

Same Prompt, Two Sessions: The Setup

Date run: October 12, 2026 Tool: Claude Code CLI 2.1.233 Model: claude-sonnet-5-5 (the same model for both sessions)

The prompt

Both sessions received exactly this text, word for word:

Write a Python script using pandas that reads ap_payments.csv (columns: payment_id, vendor_name, invoice_number, amount, payment_date) and finds possible duplicate payments: same vendor, same invoice number, same amount, paid within 30 days of each other. Write the results to duplicate_payment_exceptions.csv. Reply with only the code.

It is written the way an accountant might ask: plain English, the approved criteria stated, nothing about edge cases.

How the sessions were kept independent

Each session was a separate, one-shot run from its own empty folder:

claude -p "<prompt>" --model claude-sonnet-5-5 --no-session-persistence

• Neither session could see this repository, the article, our own script, the sample data, or the other session’s answer.
• Neither had any earlier conversation to draw on.

What was saved

• session_a.py and session_b.py are each session’s reply, unedited. The only change is removing the markdown code fences around the code.
• compare.py runs both scripts on a copy of the article’s synthetic ap_payments.csv and compares what each did with every seeded edge case against the hand-checked expected_exceptions.csv.

The results are in RESULTS.md.

Run it again yourself and you may get a third version. That is rather the point.

This site is open source. Improve this page.