READER EDITION / ISSUE 002 / Revised 2026-11-20
Creative Agentic Computing
CodeFarmer / Issue 002
YOUR OWN AI WORKSHOP

One sentence.
A different build.

By Alvin

An explicit inclusion rule gave Llama a different job. On the same M5 Max with Ollama 0.32.0 and the same model digest and options, we inserted this sentence after the notebook’s purpose:

Complete brief / keep the acceptance checks
Include every input note, whether confirmed is true or false.

The fresh function retained the unconfirmed notes. It also enclosed their quotes in brackets. It passed the empty-list fixture and failed the other three: inclusion improved while verbatim preservation broke.

if confirmed:
    report.append({'source_id': source_id,
                   'quote': quote})
else:
    report.append({'source_id': source_id,
                   'quote': f"[{quote}]"})

Give status its own place

The model marked uncertainty by changing the quotation itself. A useful app could instead display an unconfirmed badge beside the original words. That requires a separate status field and display code that leaves the quoted text intact.

The clarification produced one fresh candidate. Its exact prompt, reply and four cases are in primary-clarification-results in the lab pack. Repeat the exercise on your task to assess its reliability.