What the cohort
actually did.
Every checklist tick, exit-ticket response, session sign-off, and rubric assessment emits an xAPI statement. This page reads the statement store, rolls it up against the 25-criterion rubric bar and the 54 learning outcomes, and surfaces the three signals that drive program revision: where the cohort is stuck, where raters disagree, and which LOs aren't getting evidence.
Rubric criterion pass-rate
Proficient-floor pass rate per criterion. <70% cohort-wide signals a session gap, not a learner gap.Cohort-over-cohort comparison
Snapshot the current cohort, then diff it against a prior one. Per-criterion deltas show where the program is gaining and where it's losing ground.Take a snapshot at the end of each cohort โ it freezes the criterion pass-rates, funnel, and skills-growth into an immutable record. Diffs surface only when you have two snapshots (or one + live).
Session completion funnel
Share of cohort who have marked each session complete.LO evidence coverage
Green = cohort has produced evidence against this LO this week. Gaps here flag evaluation-plan triggers.AI touchpoint uptake
BYOK Gemini calls โ per-session usage, success rate, average latency. No institutional budget; each learner runs on their own key, so low uptake is a scaffolding signal, not a cost signal.Skills growth ยท pre vs post self-assessment
0โ4 Likert on 8 skills. Left bar = S1 pre, right bar = S12 post. Negative deltas are real: they usually mean calibration, not regression.Inter-rater reliability
Two-grader sample across D1โD5. Cohen's ฮบ tracked over time. Target ฮบ โฅ 0.70.
Live ฮบ is computed from paired reviewed statements distinguished by
context.extensions.rater. A ฮบ below 0.70 on a criterion is the signal to
re-calibrate raters against the written descriptor โ not to re-grade the cohort.