Run #771

done rubric v1 session 79467814 gemini-3.7-flash 2026-09-21 02:21

What the student sent the question as it arrived, then each reading of it

OCR of the image GOOGLE_VISION

ax + by + c = 0 = 0 में 2. रैखिक समीकरण युग्म ax + by + c a b = 4 तो इनके हल हैं यदि == 2b2C2 (A) अद्वितीय हल (B) अनन्त बहुतेरे हल (C) कोई हल नहीं (D) इनमें से कोई नहीं [AI]

The phone's own reading, which differs:

ax+ by+c =0 2, रैखिक समीकरण युग्म (x + .y+ c = 0 यदि এ=৭ तो इनके हल हैं c, এ, , C) में b, (A) अद्वितीय हल (B) अनन्त बहुतेरे हल (C) कोई हल नहीं (D) इनमें से कोई नहीं AU}

The audit's reading of the question:

ax + by + c = 0 = 0 में 2. रैखिक समीकरण युग्म ax + by + c a b = 4 तो इनके हल हैं यदि == 2b2C2 (A) अद्वितीय हल (B) अनन्त बहुतेरे हल (C) कोई हल नहीं (D) इनमें से कोई नहीं

What the student sent the question as it arrived, then each reading of it

OCR of the image GOOGLE_VISION

The phone's own reading, which differs:

The audit's reading of the question:

Review team

no_issue

Product AI

nothing found

teaching Yes · mistakes No · satisfactory Yes

Each verdict takes your own. Yours is stored against the stack's, so “where is it wrong most often” is a question with an answer.

Work is the right way up Writing is oriented upright at 0 degrees.

The small symbols visible in frames 3-8 are oriented at 0 degrees upright.

seen · 80% confident
Your view
Writing is legible at video size Writing is extremely small and faint.

In frames 3-8, only a minuscule, faint symbol is written near the bottom left, making it practically unreadable.

seen · 95% confident
Your view
The work fills the frame The camera captures a vast empty page while writing is restricted to a tiny corner.

Across frames 3-8, the vast majority of the frame is empty white space, with the minute writing tucked beside the clipboard clip.

seen · 90% confident
Your view
Nothing distracting in shot Workspace is clean with only a clipboard clip visible.

A clipboard clip on the left margin and paper surface are visible throughout frames 2-8, along with a hand shadow in frames 5-6.

seen · 90% confident
Your view
The student's problem is established No problem statement or question is captured.

No question text, prompt, or spoken problem summary is visible in any frames (frames 1-8).

seen · 100% confident
Your view
Work proceeds in followable steps No structured steps are written during the session.

From frame 3 to frame 8, the page remains static with no sequential problem-solving steps added.

seen · 95% confident
Your view
It reaches an answer No completed solution is presented.

Frames 7 and 8 end with an almost blank page and no final answer or worked-out solution.

seen · 100% confident
Your view
Something was actually written Not applicable to a camera session

measured
The concept is named before the working Not assessed

Only 36 characters of speech after cleaning (0 were the language-detection artefact) — too little to judge teaching from.

heard
Your view
Explained rather than jumped to the answer Not assessed

Only 36 characters of speech after cleaning (0 were the language-detection artefact) — too little to judge teaching from.

heard
Your view
Solved step by step, not in one leap Not assessed

Only 36 characters of speech after cleaning (0 were the language-detection artefact) — too little to judge teaching from.

heard
Your view
The session opens properly Not assessed

Only 36 characters of speech after cleaning (0 were the language-detection artefact) — too little to judge teaching from.

heard
Your view
No long dead air 0 gaps over 6.0s, 0s total (0% of the session), longest 0s

Silence is below -35.0dB — a crude measure on a phone microphone, so treat a low count as weak evidence rather than proof the tutor was talking.

measured
Your view

Observations judged, but not counted as the lab objecting

The camera holds still picture moves 65.4 on average between keyframes, 2.7 hard jumps/min

Consistent with the camera being handheld — a stand would remove this entirely. A handheld phone measured 35.1. No steady-camera session has been measured yet, so the lower band is a guess.

Not counted: Measured over 183 known-bad and 29 known-good camera sessions (2026-09-21): raised on 87% of the bad and 86% of the good, and the motion distributions are the same shape — median frame difference 25 on the bad set, 27 on the good. No threshold on this number separates them, so it is reported and not counted.

measured
Your view
Exposure is usable average brightness 149.8, 2.9% of frame very dark, 11.4% blown out

Blown-out share is reported but not judged: the subject is white paper, so a good session measures 11-24% there too.

Not counted: Same two sets: it raised 6 of 183 known-bad and 2 of 29 known-good sessions. Both too few to mean anything, and the brightness distributions overlap almost exactly. The earlier reading that it 'fires more on clean sessions' was two sessions out of twenty-nine — noise, not an inverted rule.

measured
Your view
Understanding of the concept is checked Not assessed

Only 36 characters of speech after cleaning (0 were the language-detection artefact) — too little to judge teaching from.

Not counted: Raised on 87% of known-bad and 80% of known-good sessions. The criterion is honest about why in its own `why`: the transcript has no speaker labels, so a real check and a rhetorical 'ok?' are often indistinguishable, and the bar ends up one almost no session clears. Still judged and shown, because reading it beside a recording is how it gets better.

heard
Your view

What the model returned verbatim

{
  "findings": [
    {"key": "orientation", "verdict": "ok", "headline": "Writing is oriented upright at 0 degrees.", "evidence": "The small symbols visible in frames 3-8 are oriented at 0 degrees upright.", "confidence": 0.8},
    {"key": "legibility", "verdict": "bad", "headline": "Writing is extremely small and faint.", "evidence": "In frames 3-8, only a minuscule, faint symbol is written near the bottom left, making it practically unreadable.", "confidence": 0.95},
    {"key": "framing", "verdict": "weak", "headline": "The camera captures a vast empty page while writing is restricted to a tiny corner.", "evidence": "Across frames 3-8, the vast majority of the frame is empty white space, with the minute writing tucked beside the clipboard clip.", "confidence": 0.9},
    {"key": "workspace", "verdict": "ok", "headline": "Workspace is clean with only a clipboard clip visible.", "evidence": "A clipboard clip on the left margin and paper surface are visible throughout frames 2-8, along with a hand shadow in frames 5-6.", "confidence": 0.9},
    {"key": "question_captured", "verdict": "bad", "headline": "No problem statement or question is captured.", "evidence": "No question text, prompt, or spoken problem summary is visible in any frames (frames 1-8).", "confidence": 1.0},
    {"key": "structure", "verdict": "bad", "headline": "No structured steps are written during the session.", "evidence": "From frame 3 to frame 8, the page remains static with no sequential problem-solving steps added.", "confidence": 0.95},
    {"key": "completion", "verdict": "bad", "headline": "No completed solution is presented.", "evidence": "Frames 7 and 8 end with an almost blank page and no final answer or worked-out solution.", "confidence": 1.0}
  ]
}

Duration	66s (1.1 min)
Video	960×540 @ 15.0fps
Audio track	yes
Shot changes
Dead air	0 gaps ≥6.0s · 0s total · longest 0.0s
Luma	avg 149.8 · min · max
Medium	camera 26.2% flat background
Content written	9.0% → 43.52% (+56.53)
Transcript	36 characters of speech
Frames	8

Run #771

The recording

What the student sent the question as it arrived, then each reading of it

Transcript

What the student sent the question as it arrived, then each reading of it

Transcript

Review team

Product AI

Measured

What the model returned verbatim

Frames in order, one every ~8s