Claims

Contingent tutoring held delegation at the level of an answer-on-request assistant while delivering more artifacts, one decision at a time

Claim · All claims
Evidence · 1 study (3 entries) · causal r2 · q3
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims

Reading the evidence codes

q is the design tier, i the impact, n the sample, and the kind with its r the rigour judged within that kind: what each code means.

This claim: no standardized effect is recorded in observations/ for it yet, so no benchmark reading is possible.

q3 i? Learners asked the contingent tutor to build as often as the unrestricted assistant, and far more often than the question-only tutor. → Hou 2026 q3 i? The contingent tutor delivered a tutor-written artifact in more sessions than either baseline and wrote more code per session than the assistant. → Hou 2026 (2) q3 i? Learner turn counts under the contingent tutor matched the question-only tutor and exceeded the assistant's. → Hou 2026 (3)

Evidence

Hou 2026

Hou, X., Weng, Y., Yeh, C. H., Lee, D. L., Zheng, L., Li, F., Wang, W., & Liu, Y. (2026). Adaptive Scaffolding Needs Contingency: An AI Tutor That Escalates and Fades on What the Learner Does. arXiv preprint. https://arxiv.org/abs/2609.22993 doi:10.48550/arxiv.2609.22993

q3 · i? · causal · r?

Session-level GEE analysis clustered on participant from the event log (N=131, 393 sessions). Build-request rates were 50.4% (Tutor C), 51.1% (Tutor A) and 23.7% (Tutor B); the C–A odds ratio was 0.95 and the C–B odds ratio 3.29.

"At least one build request occurred in 50.4% of Tutor C sessions and 51.1% of Tutor A sessions (𝑂𝑅= 0.95, 95% CI[0.64, 1.42],𝑝=. 82), against 23.7% of Tutor B sessions (𝑂𝑅= 3.29[1.98, 5.48],𝑝<. 001)"

Hou 2026 (2)

Hou, X., Weng, Y., Yeh, C. H., Lee, D. L., Zheng, L., Li, F., Wang, W., & Liu, Y. (2026). Adaptive Scaffolding Needs Contingency: An AI Tutor That Escalates and Fades on What the Learner Does. arXiv preprint. https://arxiv.org/abs/2609.22993. doi:10.48550/arxiv.2609.22993

q3 · i? · causal · r2

Event-log delivery counts checked against an authorship probe (κ = .76). The contingent tutor delivered artifacts in 48.1% of sessions versus 23.7% for the assistant and 9.2% for the question-only tutor, and wrote more code per session (RR = 1.64), while complete programs in chat were as frequent under Tutor A as Tutor C (RR = 1.06, p = .77).

"A tutor-written artifact was delivered into the workspace in 48.1% of Tutor C sessions, against 23.7% for Tutor A (𝑂𝑅= 2.99[1.80, 4.98],𝑝<. 001) and 9.2% for Tutor B ( 𝑂𝑅= 9.71[5.29, 17.85], 𝑝<. 001)"

Hou 2026 (3)

Hou, X., Weng, Y., Yeh, C. H., Lee, D. L., Zheng, L., Li, F., Wang, W., & Liu, Y. (2026). Adaptive Scaffolding Needs Contingency: An AI Tutor That Escalates and Fades on What the Learner Does. arXiv preprint. https://arxiv.org/abs/2609.22993. doi:10.48550/arxiv.2609.22993

q3 · i? · causal · r2

Dialogue counts per session from the event log in the same within-subjects study. Mean learner turns were 16.8 under the contingent tutor, 14.0 under the question-only tutor and 5.0 under the assistant, with rate ratios as printed.

"Learners produced a mean of 16.8 turns per session with Tutor C and 14.0 with Tutor B, statistically indistinguishable (𝑅𝑅= 1.16[0.94, 1.44],𝑝=. 17), against 5.0 with Tutor A (𝑅𝑅= 3.30 [2.72, 3.99],𝑝<. 001)"

Discussion

Page metadata
Field Value
type claim
id contingent-tutor-delegation-held-delivery-rose
title Contingent tutoring held delegation at the level of an answer-on-request assistant while delivering more artifacts, one decision at a time
description Contingent tutoring held delegation at the level of an answer-on-request assistant while delivering more artifacts, one decision at a time
status review
evidence_strength moderate
newest_evidence 2026
generated by process:wiki-ingest on 2026-10-10
trust unverified — no one has confirmed this page's content
sources 3 — hou-2026, hou-2026-2, hou-2026-3

Frontmatter is the page's machine-readable record — see the Schema & Guide.