System reliability and latency acted as a prerequisite bottleneck: 56.4% of started solves completed, with a 32.0-second average latency and 68.5-second p90¶

Claim · All claims
Evidence · 2 studies · 2 design r3 · q2
Currency · no dated study recorded · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? Students started 149 solve attempts and 84 completed successfully (56.4%), a funnel the authors treat as combining user behavior with a system-stability event. → solve-funnel q2 i? Average solve latency was 32.0 seconds with p90 at 68.5 seconds, disrupting cognitive momentum and causing premature abandonment. → solve-latency

Evidence¶

solve-funnel¶

Harry Feng, Yuan Tian, and Erica Zhao. 2026. From Answer Generators to Reasoning Facilitators: Designing AI Tutors for Mathematical Reasoning in High-Stakes Environments. https://arxiv.org/abs/2607.01692 doi:10.48550/arxiv.2607.01692

q2 · i? · design · r3

Telemetry funnel from the 12-day field deployment. The article cautions this "should not be interpreted as simple user disinterest" because a reliability breakdown on May 20–21 depressed completion.

"students started 149 solve attempts, and 84 completed successfully, yielding a 56.4% completion rate."

solve-latency¶

Harry Feng, Yuan Tian, and Erica Zhao. 2026. From Answer Generators to Reasoning Facilitators: Designing AI Tutors for Mathematical Reasoning in High-Stakes Environments. https://arxiv.org/abs/2607.01692

q2 · i? · design · r3

Latency telemetry from the same deployment; the authors state prolonged waits "directly disrupt[ed] students' cognitive momentum and caus[ed] premature abandonment."

"the average solve latency was 32.0 seconds, with the p90 reaching 68.5 seconds."

Discussion¶

Related Claims¶

• Follow-up support shows low initiation but reliable completion: the bottleneck is noticing and composing the question, not the AI's ability to answer — related
• Students rarely open transfer-practice cards immediately after solving; they want such problems organized into delayed, spaced wrong-book review instead — related
• Under a simulated load of 100 concurrent users, Taklif.AI averaged 30 seconds end-to-end response time per request — related
• On the live single-node deployment, short-answer rounds take ≈32 s (≈$0.015/round) while multiple-choice rounds are ≈4× faster (≈7.6 s, ≈$0.005/round) because grading is deterministic — related