行为 Regression 与历史 ON/OFF 比较属于不同证据;稳定科研增量收益尚未建立。 / Behavior regression is separate from capability comparisons; reliable research gain not established.
v0.2.5 Research-delivery Regression / 科研交付行为回归
v0.2.5 · gpt-6.1-sol / medium · PASS 10, FAIL 0, INCONCLUSIVE 0, NOT RUN 0; engineering assessment, no independent human adjudication
Five near-pairs A1–E2: task purpose, installation boundaries, style and actual file delivery. No OFF / Short Prompt gain test.
Can calibration evidence identify material response? / 材料响应可识别性
v0.1-alpha.2 · GPT-6.1 Sol / High · internal comparison; no substantive ON/OFF advantage established
Four-cell local affine constitutive response and calibration identifiability. / 本构响应与标定信息。
Can linear stability rule out sustained tokamak turbulence? / 线性稳定性与持续湍流
v0.1-alpha.2 · GPT-6.1 Sol / High · internal comparison tied; incremental advantage not established
Distinguishing linear decay, transient growth and finite-amplitude long-time behavior. / 线性衰减与非线性长期状态。
Can ITG gradient scans distinguish turbulent states? / ITG 湍流态判别
v0.1-alpha.2 · GPT-6.1 Sol / High · narrow positive signal; reliable incremental value not established
Coexisting attractors versus intermittency, transients and numerical selection. / 吸引态、间歇性与数值选择。