现实世界的视频基准测试提供了广泛的覆盖范围,，但其固定剪辑纠缠了事件计数, 速率, 持续时间, 和视觉复杂性, 使得故障模式难以隔离。虽然现有的程序化基准提供了更好的控制,，但它们仅对最终答案进行评分，而不是根据可执行的基本事实审核报告的事件。为了弥补这一差距,，我们引入了基于跟踪的参数分析，用于三个受控视频任务:弹跳球墙接触,视觉闪烁,和分类状态转换中的事件计数。在 2,190 个视频, 中，我们改变事件计数 N 和频率 F，同时保持渲染固定。每个视频都包含用于能力表面估计和时间戳级别评估的可执行事件跟踪。我们的结果揭示了阶段性的暂时性失败。在 80% 可靠性阈值, 下，Gemini 3.6 Flash 在 0.5 和 1.0 Hz, 下可靠地对多达 12 个事件的持续状态转换进行计数，但对于瞬态闪烁事件没有可靠的正计数区域。因此，, 事件表示决定了模型最初是否访问证据——随着计数和频率的增加，这种限制会加剧。在高计数, 高频状态, 中，最终计数中只有 0.2% 是正确的，模型仅恢复了 18.1% 的真实事件。为了测试视觉访问是否是主要瓶颈,，我们提高了采样率。尽管这将弹跳球的准确度从 19.6% 提高到 29.3%,，但报告的序列仅在 3.7% 的情况下与地面真实情况一致。因此，额外的帧数可能会夸大最终分数，而不会产生忠实的事件恢复。不同的提示策略产生的收益同样有限,，并且现实世界的视频评估显示在低事件计数下成功的集中度相同。最终，, 基于跟踪的分析将视频评估从聚合准确度指标转变为对时间推理失败的位置进行详细诊断。

Real-world video benchmarks provide broad coverage, but their fixed clips entangle event count, rate, duration, and visual complexity, making failure modes hard to isolate. While existing programmatic benchmarks offer better control, they score only the final answer rather than auditing reported events against executable ground truth. To bridge this gap, we introduce trace-grounded parametric profiling for event counting in three controlled video tasks: bouncing-ball wall contacts, visual blinks, and categorical state transitions. Across 2,190 videos, we vary event count N and frequency F while holding rendering fixed. Each video includes an executable event trace for capability-surface estimation and timestamp-level evaluation. Our results reveal a staged temporal failure. At an 80% reliability threshold, Gemini 3.6 Flash reliably counts persistent state transitions up to 12 events at 0.5 and 1.0 Hz, yet demonstrates no reliable positive-count region for transient blinking events. Thus, event representation dictates whether a model initially accesses evidence -- a limitation that compounds as count and frequency increase. In the high-count, high-frequency regime, only 0.2% of final counts are correct and the model recovers just 18.1% of true events. To test if visual access is the primary bottleneck, we increase sampling rate. Although this boosts Bounce Ball accuracy from 19.6% to 29.3%, the reported sequence agrees with ground truth only 3.7% of the time. Extra frames can therefore inflate final scores without producing faithful event recovery. Different prompting strategies yield similarly limited gains, and real-world video evaluations show the same concentration of success at low event counts. Ultimately, trace-grounded profiling shifts video evaluation from aggregate accuracy metrics to a detailed diagnostic of where temporal reasoning fails.

科目:人工智能(cs.AI)

Subjects: Artificial Intelligence (cs.AI)