Ten models refreshed with 322 complete outputs

Explore gemma-4-26b-a4b-it, qwen-2.5-7b-instruct, nova-lite-v1, qwen3.5-flash-02-23, qwen3-coder-30b-a3b-instruct, mimo-v2.5, qwen3-14b, qwen3-32b, seed-1.6-flash, gpt-oss-safeguard-20b.

Of 350 attempted runs, 26 ended with model or provider failures and two failed during final capture. Those 28 runs are excluded from this publication; their previously published data remains unchanged. Saved outcomes are retained for investigation.