Back to Agno

Test Log - _18_quality_review

cookbook/data_labeling/_18_quality_review/TEST_LOG.md

2.8.01.4 KB
Original Source

Test Log - _18_quality_review

Tested 2026-07-18 against gemini-3.5-flash (labeler A) and claude-opus-4-7 (labeler B, reviewer, adjudicator), agno 2.7.4.

basic.py

Status: PASS

Description: Labeler / reviewer / adjudicator pipeline expressed as an agno Workflow (Parallel labelers, reviewer, Condition-gated adjudicator) on a Contact extraction task. The demo runs two inputs - a clean one and one with deliberately conflicting details - and prints the full step trail for both, exercising both the skip path and the adjudication path.

Result: Clean input: both labelers returned identical Contact objects (name='Liam Ortega', email='[email protected]', company='Meadow', title='Support Engineer'), the reviewer returned needs_adjudication=False, and the trail shows the Condition skip ('Condition Adjudicate not met - skipped 1 steps'). Conflicting input: the labelers disagreed on name ('Dr. Sarah Chen-Watanabe' vs 'Sarah Chen') and title ('Principal Scientist and acting Head of Platform' vs 'Principal Scientist'); the reviewer flagged both fields with per-field reasons; the adjudicator ran and emitted FinalLabel(contact=Contact(name='Sarah Chen', email='[email protected]', phone='+1-415-555-0177', company='NovaLabs', title='Principal Scientist and acting Head of Platform')). Which fields disagree can vary run to run; the conflicting input is constructed so that at least one field reliably does.