What did each programme's graduates actually apply for, and actually accept?
The question. For each MSB programme and graduating class, which job titles and employers did students apply to, and which did they accept?
Withheld
NOT YET BUILT, AND LICENCE-BOUND WHEN IT IS. Career Services placement data carries the same internal-use-only licence as the Handshake import, so core.placement_facts gets no anon grant and renders only behind prog membership. Derived aggregates - a rule, a prior, a crosswalk weight - are publishable because they are no longer the licensed rows. Needs the programme data request to go out before anything can be ingested.
This analysis is listed rather than hidden so that a reader can tell the difference between a question we are not answering yet and one we never asked.
What it found
The observed-fact anchor the rubric currently lacks, and an input to three analyses rather than a page of its own. Proposes core.placement_facts keyed (programme, cohort, title_norm, employer, disposition), where disposition is applied / interviewed / offered / accepted. Accepted rows carry assertion_status = observed, which outranks model_inference in the existing enum, so a Career Services record beats the classifier by construction rather than by argument. Today only one programme exists in any form: 75 Marketing titles for 98 graduates, class of 2026, in core.seniority_gold.
How it was computed
One annual file per programme per class, normalised with the same title stripper the classifier uses, employers resolved against core.employer_canonical, upserted on the unique key so a resent file is idempotent. The rubric loop runs four queries after each ingest: panel coverage, classifier-versus-placement disagreement, the prior check, and promotion. A disagreement becomes a rubric rule only at 3 or more graduates across 2 or more cohorts, and the resulting rule is re-scored against the held-out seed-20260915 gold set, never the set it was derived from.
How to read it
Applications are not acceptances and must never be summed together; only the accepted rows are observed fact, applications are evidence about student belief. The promotion threshold will be most tempting to lower in the first year, when the evidence is thinnest. The single-programme figures already in hand are suggestive, not representative: 49.0% of Marketing graduates took a title that appears nowhere in the 295,354-posting panel.
Source view: none