| Deterministic debug-loop fixture can return a later scripted result after an extra tool call. |
agentic artifacts and ticket |
Open product-fixture defect; add an extra-call regression before using this scenario as a strict model gate. |
| Qwen agentic outcomes are internally above the suite's 0.75 floor but below the campaign's 1.0 gate. |
run plan, greedy scout, sampling scout |
Preserve both thresholds; do not relabel the native artifacts as campaign passes. |
| Ordered planning terms can fail when an incidental later-term substring appears in an earlier section. |
paired GLM scout and ticket |
Open scorer defect; add positive and negative heading-order regressions before comparative agentic ranking. |