AI

Labels Override Definitions in Jev-Style Typed Decision Models

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 06/10/2026, 19:18:24 UTC.

Briefings

A study of Jev-style typed decision models reveals that labels often override written definitions, causing option-label bias. Using four open-weight models, researchers found that accuracy remains stable when definitions are deleted, but accuracy drops significantly when labels contradict their definitions. The findings highlight a specific failure in prompt rendering within the decision head, with no prior background provided.

Key points

• Accuracy remains stable at 0.8487 when definitions are deleted, but drops to 0.2281 when labels contradict their definitions in the von model.
• Renaming options to A and B increases accuracy by +0.1511, demonstrating that label naming significantly impacts model performance.

Context

This research investigates Jev-style typed decision models, where developers define rules via labels and definitions. The study analyzes open implementations to determine if model behavior aligns with stated definitions or label preferences.

Limitations

Evidence is limited to four open-weight models and one synthetic routing suite. Results are based on a single snapshot from September 2026, and the study does not provide additional background context or causal claims beyond the observed label-definition conflict.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 23/10/2026, 08:58:31 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeibg6jnmgufu3lr4tdyhnixn6zokg3okknjdl5klcfrazbfpzwj4aq