Public & anonymized. Same page for everyone, no account needed. Rows hold shape, never identity: no ip, no cookie id, no user id, no answer text. Cells under 7 observations stay hidden. How this data works · JSON
nu (pre-K through grade 12) · Last 28 days · computed 2026-11-12 20:36 UTC · refreshes every few minutes
273Model & service calls
98.5%Succeeded
1sMedian latency
2.17Calls per graded answergrading purposes only
66697Characters sent
1791Characters received
Every outbound call to a language model, the decision classifier, translation, speech, lesson generation or code execution. Purpose, host, latency & outcome only. No prompt or completion text.
By purpose
Purpose
Calls
Share
Succeeded
Median
95th
Chars in
Chars out
Text to speech
253
92.7%
98.8%
1s
1s
31231
0
Classification (chat model)
7
2.6%
85.7%
1s
10s
25943
249
Code execution
6
2.2%
100.0%
1s
1s
65
0
Feedback
6
2.2%
100.0%
1s
2s
9458
1516
Speech to text (mic)
1
0.4%
100.0%
0s
0s
0
26
By lesson
Which lessons drove model calls — chat, feedback and speech attributed to the lesson the learner was on.