WEBVTT

1
00:00:00.220 --> 00:00:04.760
Consider a neutral timeline with Phase A, Event B, and Phase C.
先看中性的 A 阶段、B 事件和 C 阶段时间线。

2
00:00:05.440 --> 00:00:11.440
The question asks what happened immediately before Event B, so
finding B is only the first step.
问题问 B 事件之前发生了什么，找到 B 只是第一步。

3
00:00:12.240 --> 00:00:17.540
Inspecting only the frames inside B may miss the earlier
interval that actually contains the answer.
只检查 B 内部画面，可能错过真正包含答案的较早区间。

4
00:00:18.200 --> 00:00:24.080
The system needs to localize the event, inspect the preceding
window, and connect the evidence across time.
系统需要定位事件、检查前一时间窗，并关联跨时间证据。

5
00:00:24.940 --> 00:00:30.320
This original diagram contains no patient details, operative
images, or recommended medical actions.
原创图不含患者信息、手术影像或医疗操作建议。

6
00:00:30.880 --> 00:00:35.380
It explains the retrieval problem MedClaw studies in long
surgical-video question answering.
它解释 MedClaw 在长手术视频问答中研究的检索问题。

7
00:00:36.027 --> 00:00:39.527
MedClaw separates retrieval planning from visual perception,
MedClaw 分离检索规划与视觉感知，

8
00:00:40.067 --> 00:00:44.227
using a text-only orchestrator and frozen vision-language
sub-agents.
使用纯文本协调器和冻结的视觉语言子模型。

9
00:00:45.087 --> 00:00:50.227
The orchestrator never sees the video directly; it decides which
evidence to request through tools.
协调器不直接看视频，而是通过工具决定请求哪些证据。

10
00:00:51.047 --> 00:00:57.067
Visual sub-agents inspect the specified pixels and return
textual observations that the orchestrator can reason about.
视觉子模型检查指定像素，返回供协调器推理的文字观察。

11
00:00:57.927 --> 00:01:01.807
No model weights are updated during this harness's inference or
heuristic skill-learning loop.
该框架的推理与启发式技能学习循环都不更新模型权重。

12
00:01:02.527 --> 00:01:08.467
Its adjustable part is the external library of retrieval
strategies, not a newly fine-tuned visual backbone.
可调整的是外部检索策略库，不是新微调的视觉骨干。

13
00:01:09.267 --> 00:01:14.467
Keeping those roles separate turns an implicit attention
decision into a visible request for evidence.
分离职责把隐含注意选择变成可见的证据请求。

14
00:01:14.873 --> 00:01:22.013
The tools expose progressively finer evidence: a full recording,
a chosen temporal segment, and individual frames.
工具逐步提供更精细证据：完整记录、选定时间段和单帧。

15
00:01:22.873 --> 00:01:28.473
A full-video pass produces a rough timeline and then revisits a
candidate window at higher resolution.
完整视频读取先生成粗时间线，再以较高分辨率复看候选窗口。

16
00:01:29.273 --> 00:01:34.613
A segment tool reads the localized interval or an adjacent
interval needed for a temporal question.
片段工具读取定位区间，或时间问题所需的相邻区间。

17
00:01:35.593 --> 00:01:41.753
Independent frame inspection can check a specific visible
detail, including a smaller spatial crop when needed.
独立单帧检查可核对可见细节，必要时检查更小空间区域。

18
00:01:42.613 --> 00:01:47.292
A knowledge-retrieval tool supplies general background terms,
rather than answers from the benchmark.
知识检索工具提供一般背景术语，不提供基准答案。

19
00:01:48.053 --> 00:01:53.453
Background knowledge and the evidence from this particular
recording remain different inputs to the reasoning process.
背景知识与该记录本身的证据，是推理中的不同输入。

20
00:01:54.053 --> 00:02:00.073
Each tool call names its arguments, such as a time range, and
returns an observation into the running context.
每次工具调用明确时间范围等参数，并把观察加入当前上下文。

21
00:02:00.893 --> 00:02:05.513
The orchestrator then decides whether the evidence is sufficient
or another look is needed.
协调器随后判断证据是否充足，或是否需要再看。

22
00:02:06.413 --> 00:02:11.493
The action-observation sequence is recorded until the model
stops and produces its final answer.
操作与观察序列被记录，直到模型停止并生成最终答案。

23
00:02:12.353 --> 00:02:18.133
In the original before-B example, an event-window observation
should trigger inspection of the preceding interval.
在原创 B 之前示例中，事件窗口观察应触发前一区间检查。

24
00:02:18.673 --> 00:02:24.293
That makes the requested re-look visible, instead of hiding the
retrieval decision inside one compressed pass.
这样复看请求可见，不把检索决定藏在一次压缩读取里。

25
00:02:25.393 --> 00:02:31.573
A recorded trace supports inspection, but recording a decision
alone does not guarantee its interpretation is correct.
轨迹支持检查，但记录决定本身不保证解释正确。

26
00:02:32.180 --> 00:02:37.500
Heuristic Skill Distillation starts from one hundred labeled
questions in the separate SVU training split.
启发式技能蒸馏从独立 SVU 训练划分的一百个标注问题开始。

27
00:02:38.140 --> 00:02:44.140
Sixty questions support candidate mining, and forty form the
validation set used to gate those candidates.
六十个问题用于挖掘候选，四十个用于验证门控。

28
00:02:44.880 --> 00:02:50.700
The agent first answers blind, then compares its answer and
retrieval trace against the reference during mining.
挖掘时智能体先盲答，再将答案和检索轨迹与参考比较。

29
00:02:51.540 --> 00:02:58.640
An operational miss can suggest a tighter crop or adjacent
window; an unreachable answer offers no retrieval lesson.
操作性遗漏可提示更紧裁剪或相邻窗口；不可获取答案不提供检索教训。

30
00:02:59.320 --> 00:03:05.600
A skill must beat the empty-library mean by over zero point zero
five, with no answer losing two or more points.
技能须比空技能库平均分高超过零点零五，且单题不能下降两分或更多。

31
00:03:06.340 --> 00:03:11.340
The retained lesson changes the external skill library, while
the model parameters remain frozen.
保留教训改变外部技能库，模型参数仍冻结。

32
00:03:11.927 --> 00:03:15.147
The learned library prominently includes directed re-look,
学习库重点包含定向复看，

33
00:03:15.667 --> 00:03:19.647
a strategy triggered by questions about what precedes or follows
an event.
这是由事件前后问题触发的策略。

34
00:03:20.327 --> 00:03:25.207
After localizing the event, the agent inspects the following
window for an after question.
定位事件后，智能体为后续问题检查后一个窗口。

35
00:03:25.667 --> 00:03:30.447
For a before or preparation question, it returns to the
preceding window instead.
对于之前或准备问题，它转向前一个窗口。

36
00:03:31.327 --> 00:03:33.967
The strategy addresses a specific retrieval failure:
策略针对一种特定检索失败：

37
00:03:34.507 --> 00:03:38.227
reading the named event without reading the context the question
asks about.
读了命名事件，却没读问题所需上下文。

38
00:03:38.967 --> 00:03:44.167
Our neutral timeline illustrates the idea; its phases are not
observations from a real surgical recording.
中性时间线只示意思想，各阶段并非真实手术记录的观察。

39
00:03:44.907 --> 00:03:50.887
The mechanism is evidence seeking across time, rather than
assuming the customary next step from background knowledge.
机制是跨时间寻找证据，而非依据背景知识假设惯常下一步。

40
00:03:51.553 --> 00:03:56.133
MedClawBench combines two hundred fifty-nine long neurosurgery
questions
MedClawBench 包含二百五十九个长神经外科问题，

41
00:03:56.613 --> 00:03:59.953
with eight hundred sixty-four held-out lecture-video questions.
以及八百六十四个留出的讲座视频问题。

42
00:04:00.733 --> 00:04:03.693
The total is one thousand one hundred twenty-three questions,
总计一千一百二十三个问题，

43
00:04:04.133 --> 00:04:07.233
with filtering designed to remove specified answer shortcuts.
过滤针对特定答题捷径。

44
00:04:08.193 --> 00:04:13.113
Blind and local evidence tests check whether a question requires
watching its targeted interval.
盲答和局部证据检查判断问题是否需要观看目标区间。

45
00:04:14.133 --> 00:04:20.053
Lecture questions also face an audio-only filter, reducing
answers obtainable by listening to narration alone.
讲座问题也经过仅音频过滤，减少只听解说就能答的问题。

46
00:04:21.253 --> 00:04:23.973
Answers receive four one-to-five rubric scores:
答案接受四项一至五分规则评分：

47
00:04:24.553 --> 00:04:29.353
correctness, detail, contextual understanding, and temporal
understanding.
正确性、细节、上下文理解和时序理解。

48
00:04:30.373 --> 00:04:32.953
These benchmark answer scores are research measurements,
这些基准问答分数是研究测量，

49
00:04:33.293 --> 00:04:36.433
not patient outcomes or a test of clinical deployment readiness.
不是患者结局或临床部署准备度测试。

50
00:04:36.940 --> 00:04:43.060
A controlled ablation wraps the same Qwen two point five V L
seven-billion-parameter visual backbone in the harness.
受控消融把相同七十亿参数 Qwen2.5-VL 视觉骨干放进框架。

51
00:04:43.900 --> 00:04:48.860
On the neurosurgery split, its one-shot four-dimensional average
is two point zero four.
在神经外科划分上，单次读取的四维平均分为二点零四。

52
00:04:50.160 --> 00:04:53.860
With the harness, the same visual-backbone row reports two point
five eight.
加入框架后，相同视觉骨干行报告二点五八。

53
00:04:55.000 --> 00:04:59.900
The added orchestrator and tool loop are part of the comparison;
the visual model itself is not retrained.
比较包含新增协调器和工具循环，但视觉模型本身不重训。

54
00:05:01.120 --> 00:05:06.580
A separate component ablation reports a zero point three eight
score reduction when directed re-look is removed.
另一组件消融报告移除定向复看后分数降低零点三八。

55
00:05:07.600 --> 00:05:13.340
Read these as scoped rubric-score improvements, not
probabilities that an agent can safely interpret any clinical
video.
这些是限定范围的规则分提升，不是安全理解任意临床视频的概率。

56
00:05:13.960 --> 00:05:19.760
Return to Event B: locating the event is useful, but the
question may point to another nearby interval.
回到 B 事件：定位很有用，但问题可能指向邻近区间。

57
00:05:20.340 --> 00:05:25.580
A text planner can choose that interval, while a visual tool
inspects what the recording actually contains.
文本规划器选择区间，视觉工具检查记录实际包含的内容。

58
00:05:26.340 --> 00:05:30.420
The observations and decisions remain in a trace that can be
inspected and revisited.
观察和决定留在可检查、可复看的轨迹中。

59
00:05:31.320 --> 00:05:36.320
A validated retrieval lesson can later guide the next question
without changing the underlying model weights.
经验证检索教训可指导后续问题，而不改变底层模型权重。

60
00:05:37.300 --> 00:05:40.920
MedClaw studies that harness design under a defined long-video
benchmark,
MedClaw 在明确的长视频基准下研究该框架，

61
00:05:41.300 --> 00:05:43.400
with evidence gathering as the central mechanism.
其核心机制是证据收集。

62
00:05:44.260 --> 00:05:49.040
The idea is to improve how a system looks for support before
trusting the answer it produces.
其思想是在信任答案之前，先改善系统寻找依据的方式。