The BaseResearcharXiv

Known Good, Known Bad: A Ground-Truth Calibration Protocol for AI Judgment in Science

According to the abstract, Olga Lavinda proposes a protocol that tests AI judgment against a verified reference case and deviant variants, and demonstrates it with an organic chemistry example in which a free ChatGPT model detected three of five intended deviations at first.

Open the source ↗

Why it matters

The same thing, for different readers.

What it means for you

An AI chatbot may miss planted errors in a calculation until pushed to recheck, so its first answer should not be taken as verified.

For researchers

Olga Lavinda offers a small, descriptive protocol; six conversations from one unidentified backend cannot support general claims.

How the finding was reached

In our words, from the source. Weigh the finding by how it was made.

What was done
An instructor-generated organic chemistry demonstration: six conversations and 18 responses from an unidentified backend on the free logged-out ChatGPT interface in July 2026.
What it found
Neutral evaluation detected three of five intended deviations; both initial misses were detected at the authority-framed second turn, and all five persisted at the third-turn audit.
What it does not show
Single demonstration, small sample, unidentified backend; described by the author as descriptive.
Status
A preprint: not yet peer reviewed

The record

Questions
How is AI changing childhood, health and daily life?
People
Olga Lavinda (Wrote it)
What we read
We have read only the description the publisher gives. No quotations until someone reads the source itself.
Found by
The Radar, run of November 8, 2026, via arxiv-ai-safety-query (tier 1: the source’s own channel or an official record)
Cite this
arXiv, “Known Good, Known Bad: A Ground-Truth Calibration Protocol for AI Judgment in Science”, November 8, 2026. Entry in the Ctrl AI Base, https://ctrlai.com/base/arxiv-2026-11-08-known-good-known-bad-calibration
Our notesLight lane: written from the triage excerpt only.