Computer science curriculum
Grounding is not a product feature, it is a curricular concept. For people deciding what belongs in a syllabus rather than which tool to adopt.
What the session covers
A student who understands why an agent must cite a dataset ID understands provenance, API contracts, and verification — three things that outlive whichever model is current. NYC’s open data portal is an unusually good teaching corpus because the data is real, local, free, and messy in instructive ways.
- The failure mode, demonstrated not asserted.
- The catalog is an API, and that is the lesson. Step one returns a dataset ID and a column list. That is a contract. Step two queries against it. The two-step generalizes far past this portal.
- The CS pipeline in the city’s own data — enrollment and outcomes, with the caveats the data itself carries.
- When the finding is the absence, which is the hardest move to teach and the one student writing most often skips.
- What it takes to put this in a classroom.
What it is honest about
Every prompt was dry-run against the live connectors on 2026-07-21, and the presenter notes record what each one actually returned — including the ones that returned less than hoped.
No deck for this one yet. The demo script is complete and dry-run; the slide companion has not been built. Its sibling — liberal arts and general education — has one, and covers the same arc for a non-CS audience.