Marble's age bands are authored priors (start-age distribution comes in waves: 256 topics at 5, 289 at 7, 308 at 9). Instrumenting readiness curves per topic (P(mastery | age, prereqs-met)) is nearly free given the calibration plan, and divergences are product improvement + upstream contribution + publishable learning-progressions research.
Decide: (a) pre-register the research intent (cheap, forces rigor, raises credibility)? (b) IRB/ethics path if we publish beyond aggregate calibration; (c) guard against selection bias — early-adopter kids ≠ population; findings stay framed as "our-population calibration" per ADR 0003 rule 3.
Context: 06-cognitive-science.md, Q6.2
Marble's age bands are authored priors (start-age distribution comes in waves: 256 topics at 5, 289 at 7, 308 at 9). Instrumenting readiness curves per topic (P(mastery | age, prereqs-met)) is nearly free given the calibration plan, and divergences are product improvement + upstream contribution + publishable learning-progressions research.
Decide: (a) pre-register the research intent (cheap, forces rigor, raises credibility)? (b) IRB/ethics path if we publish beyond aggregate calibration; (c) guard against selection bias — early-adopter kids ≠ population; findings stay framed as "our-population calibration" per ADR 0003 rule 3.
Context: 06-cognitive-science.md, Q6.2