WDCD Comparative Review: Security Compliance Lowest at 1.15, 2.85-Point Compliance Gap Across 11 Models

In the WDCD v3.1 five-scenario comparative review, the security compliance scenario proved to be the weakest link in instruction-following capability across all models. qwen3-max ranked last with just 1.15/4, while gpt-5.5 led with 4/4, resulting in a maximum within-scenario score gap of 2.85 points.

Security Compliance Scenario: The Defense Most Easily Breached Under Pressure Rounds

The security compliance scenario showed the most dispersed score distribution, with a 2.85-point gap between last-place qwen3-max (1.15/4) and champion gpt-5.5 (4/4). doubao-pro likewise managed only 1.7/4, and gemini-2.5-pro scored 2.25/4. In the v3 item design, the consecutive pressure phase includes two rounds: authority special approval and salami-slicing. When constraints touch compliance boundaries, models are more likely to break down in rounds 6-8. By contrast, the lowest score in the data boundary scenario was deepseek-v4-pro at 2.8/4, a gap of only 1.2 points, indicating that hard compliance constraints are the most difficult to sustain under multi-round incremental pressure.

Engineering Standards Scenario: Zero Differentiation Exposes Testing Ceiling

All 11 models scored a perfect 4/4 in the engineering standards scenario, making it the only scenario with no differentiation. In the three-round design of the v2 anchor items, every model held its constraints during the R3 pressure stage, with all S_hold scores at full marks. This shows that the current item pool does not exert enough constraint intensity on engineering standards to differentiate model capabilities. Enterprises relying solely on this scenario for model selection will fail to obtain effective signals.

Resource Constraints and Business Rules: Lopsided Models Come to Light

In the resource constraints scenario, claude-opus-4.7, deepseek-v4-pro, and gpt-o3 all scored 4/4, while doubao-pro managed only 2.35/4 and claude-sonnet-4.6 scored 2.6/4. The business rules scenario was led by glm-4.6 at 4/4, with claude-sonnet-4.6 at the bottom with 2.25/4. Among the lopsided models, doubao-pro showed a 2.3-point gap between engineering standards (4/4) and security compliance (1.7/4), and gemini-2.5-pro showed a 1.75-point gap between data boundary (4/4) and security compliance (2.25/4). These gaps stem from differences in memory retention rates across constraint types during the KBV recitation probe stage of the v3 items.

Practical Implications for Enterprise Production Workflow Integration

For enterprises integrating AI into production workflows, models scoring below 2.5/4 in the security compliance scenario must be equipped with mandatory guardrails. The performance of qwen3-max and doubao-pro in this scenario means that in workflows involving data masking and permission approval, relying solely on model compliance carries excessive risk. gpt-5.5 and gpt-o3, both at 4/4 in security compliance, can be prioritized for compliance-sensitive processes. For models scoring below 3/4 in the resource constraints scenario, it is advisable to add external verification in budget control and concurrency limit tasks.

Strategic Assessment: Overvalued and Undervalued Models

claude-sonnet-4.6 scored 4/4 in engineering standards but only 2.25/4 in business rules, a 1.75-point gap, suggesting that its compliance capability may be overvalued by the market, particularly in business processes requiring parallel hard constraints. qwen3-max's 2.85-point gap between engineering standards (4/4) and security compliance (1.15/4) highlights a vulnerability in compliance scenarios that warrants focused validation in the next evaluation cycle. glm-4.6, with 4/4 in both business rules and data boundary, has its overall compliance performance undervalued and can serve as a candidate for mixed multi-scenario deployment.

The perfect scores across all models in the engineering standards scenario indicate that the current v3.1 item pool has reached its ceiling in pressure intensity for this constraint. The next cycle needs to add compound pressure rounds combining sunk cost and social proof to open up differentiation. The concentration of low scores in the security compliance scenario reminds enterprises that this scenario's score must be used as a primary screening criterion in model selection, rather than merely referencing the overall average.

The 1.15 security compliance score is not an isolated data point; it is the gap most likely to break first in a production environment.

Data source: YZ Index WDCD Compliance Leaderboard | Run #285 · Scenario Matrix | Evaluation Methodology