cookbook/data_labeling/_01_text_classification/TEST_LOG.md
Tested 2026-07-18 against gemini-3.5-flash, agno 2.7.4.
Status: PASS
Description: Sentiment classification (positive/negative/neutral) over three product reviews using an output_schema with a single Literal label field.
Result: All three samples classified as expected: "I love this product, fantastic quality and fast shipping." -> positive; "Broken on arrival, total waste of money." -> negative; "It works as described, nothing special." -> neutral.
Status: PASS
Description: Same task as basic.py with an extra confidence field (high/medium/low) on the output, to support routing low-confidence labels to a human queue.
Result: Confidence tracked ambiguity as intended: "Best purchase of my life, life-changing!" -> positive/high; "It's fine I guess." -> neutral/medium; "Yeah right, this thing is 'amazing'." (sarcasm) -> negative/low.
Status: PASS
Description: Same task with a free-text rationale field alongside each label; instructions ask the model to quote or paraphrase the deciding words.
Result: Both labels correct with rationales citing the deciding phrases: "Shipping was fast but the product itself fell apart in a week." -> negative ("...the product quickly fell apart within a week"); "Better than expected, will buy again." -> positive (quotes 'Better than expected' and 'will buy again').