Calificación backfill — teacher-tendency model (deferred, data-gated)
Backfill numeric calificación on ~1740 pre-MARS-215 ungraded prácticas. Locked a=8/b=3/c=7/e1=3; d(53) HELD for manual review; NULL-only (102 teacher grades untouched); pending buckets (352) excluded. Flat mapping deferred because the already-graded sample is 84% one grader (not representative) — instead build a per-teacher×bucket median model with a fallback ladder (teacher-bucket n>=threshold -> cohort bucket median -> flat). TRIGGER: execute on coverage (enough multi-grader teacher-bucket cells clear threshold), not a calendar date; re-pull distribution periodically. Full plan + data snapshot: mars/pm/grading-backfill-teacher-tendency.md (pushed 80d9748). Mechanism: Class-A migration by db-mars-cc, setArchiveContext + terminal assertions, audit design-ping, writes latest practicaReviews.calificacion WHERE calificacion IS NULL. Role: db (author migration) + coder (analysis).
Sub-tickets — 0/1 done
Questions
Activity
-
status=deferred
-
Grading-indicator view family LIVE (migration 070, SHA f397ab4, audit PASS:f397ab4). vPracticaVerdictPath (currentState-keyed SSOT) + vGradingIndicators/vGraderLeaderboard/vGradeDistribution. Corrected split: old e2=86 = 54 waiting-on-student + 32 resubmitted-waiting-on-teacher. DB-only artifacts (schema.md + reference-data.tsv exported).