L字になっている部分をどう仕上げるか。
Audio not yet verified
Evidence workspace
See what the video supports—and what still needs asking
Practitioner speech populates the judgment record; visible observations and vision-model outputs remain in a separate visual-evidence layer. They can anchor precise questions without inventing meaning or correctness.
Episode, record and probe model available
15
Speech + on-screen text
08
Direct visual observations
539
Visual annotations + derived cues
20
Real model artifacts
07
Interview probes
0:00 / 8:04
Source video
【町工場】ヤスリがけのコツ Part3【Milling machine】
家内事業町工場の日々 · 8:04
Evidence contract
Only speech populates the judgment recordVisual observations propose the next questions
Structured record
Practitioner judgment record
Speech-backed only01
Context
02
Signals
面を感じながら掛ける。
Audio not yet verified
二方向からまっすぐ掛けると、角のところがきれいに掛からない。
Audio not yet verified
03
Criteria
角のところまでヤスリが掛かっている。
Audio not yet verified
角に掛け残しがない。
Audio not yet verified
04
Action
45度で当てて掛ける。
Audio not yet verified
掛け残しがないように掛け、二つの面をつなげる。
Audio not yet verified
ヤスリになっていない側を隣の面に当てて掛ける。
Audio not yet verified
05
Reasoning
ヤスリになっていない側を当てると、隣の側を傷つけず、掛けたい側だけを削れるため。
Audio not yet verified
06
Exceptions
Not established in the source speech
02 · MULTIMODAL EVIDENCE
Visual evidence and machine observations
This separates literal video observations, authored attention, OpenCV-derived motion, and artifacts from models that actually ran. None of these alone establishes expert intent, rationale, quality, or safety.
8Direct visual observations
539Visual annotations + derived cues
20Real model artifacts
Observations and model outputs support what is visible. What it means must be checked with a practitioner or separately cited existing knowledge.
01
Direct observations from the video
Literal descriptions of visible objects, orientation, possible contact, and movement only.
02
Timecoded visual annotations and derived cues
Attention points, motion trails, hand landmarks, and related candidates. Play any card at its source moment.
03
Artifacts from visual models that actually ran
OpenAI · recorded-run
observations.jsonGoogle · mediapipe-0.10.35:fbc2a30080c3
hand-clip-01.jsonMeta · facebookresearch/sam2@2b90b9f5ceec907a1c18123530e92e794ad901a4
keyframes.jsoncurve-file-mask.pngcurve-support-glove-mask.pngcurve-workpiece-mask.pngl-angle-file-mask.pngl-angle-support-glove-mask.pngl-angle-workpiece-mask.pngstraight-file-mask.pngstraight-support-glove-mask.pngstraight-workpiece-mask.pngtool-show-glove-mask.pngtool-show-hand-tool-mask.pngtool-show-workpiece-mask.pngMeta · facebookresearch/sam2@2b90b9f5ceec907a1c18123530e92e794ad901a4
262000-268000-4fps.json262000-272000-4fps.json376000-394000-4fps.json416000-426000-4fps.json84000-94000-4fps.jsonContinue with this source
One source, five connected views
Switching views preserves the same source hash across instruction, analysis, skill structure, evidence, and human review.
01Learning guideLearn the demonstration as a visual procedure02Vision labInspect model outputs, live graph, and teaching visuals03Skill mapExplore process, tools, judgment, and learning
04EvidenceReturn to speech, source text, and observationsYou are here
05Open questionsAnswer knowledge gaps and update the skill graph