We benchmark three supervised fine-tuned models against frontier zero-shot baselines on a 661-row held-out slice of PiSAR (Persona, intent, Screen, Action, Rationale), a 12,929-tuple corpus of screen-anchored behavioural rationales curated from public app-store reviews, Pew American Trends Panel demographics, and the OPeRA shopper traces. Every model, frontier or fine-tuned, is evaluated on the same 661-row slice with the same scoring pipeline. Two findings. First, frontier zero-shot baselines (Claude Opus 4.7 and GPT-5.5) reach sem_sim 0.459 and 0.482 respectively; a fine-tuned Qwen3-VL-8B-Instruct reaches 0.783 and clears sem_sim >= 0.7 on 79% of rows, against 1-2% for either frontier baseline, a gap of 0.30 absolute on the same test set. Second, the same training data and recipe on Gemma-4-26B-A4B-IT scores only 0.441, in the same band as the frontier zero-shot baselines rather than the fine-tuned Qwen. We read this as a recipe-vs-model mismatch: the reasoning-tuned high-parameter model resists displacement and would likely need either more data or a stronger fine-tuning method.

翻译：我们对三种监督微调模型与前卫零样本基线模型进行了基准测试，使用的数据集是PiSAR（人格、意图、屏幕、动作、推理）中一个包含661行的保留子集。PiSAR是一个包含12,929个元组的屏幕锚定行为推理语料库，该语料库来源于公开应用商店评论、皮尤美国趋势面板人口统计数据以及OPeRA购物者轨迹数据。所有模型，无论是前卫模型还是微调模型，都在相同的661行子集上使用相同的评分流程进行评估。主要发现有两项。首先，前卫零样本基线模型（Claude Opus 4.7和GPT-5.5）的语义相似度分别达到0.459和0.482；而经过微调的Qwen3-VL-8B-Instruct模型达到0.783，且在79%的行上语义相似度≥0.7，相比之下，两个前卫基线模型仅达到1-2%，在同一测试集上的绝对差距为0.30。其次，使用相同的训练数据和配方对Gemma-4-26B-A4B-IT模型进行训练，其得分仅为0.441，与前卫零样本基线模型的分数区间相同，而非微调Qwen模型的水平。我们将此解读为配方与模型之间的不匹配：经过推理调整的高参数模型抵抗位移，可能需要更多数据或更强的微调方法。