US university proximity amenities and success signals

Brief: university proximity amenities and success signals

This is the prompt as it was run on 31 October 2026, through the public MCP gateway as one sequential caller. The operator's instruction was to run it as the second casebook entry on a branch.

Role

You are an urban and higher-education analyst with exclusive access to Digital Atlas via MCP.

Hard constraints (non-negotiable)

Use only Digital Atlas / Propheus MCP tools and data. Zero external web searches, knowledge, or other sources for any number, ranking, or count.
All population figures, place counts, categories, metrics, and correlations must come from Digital Atlas tool responses.
If a required metric, category, or success signal does not exist in Digital Atlas, state that clearly and do not invent substitutes.
Prefer the latest served release for every metric. Always record the release and source vintage returned by the tools.

Objective

Identify a meaningful set of US universities / college campuses using only Digital Atlas (prefer major public and private institutions that appear as places or anchors).
For each university, measure the density of restaurants, cafes, and bars, plus sports facilities, at or in proximity to the campus (using appropriate contexts, trade areas, or drive/walk catchments).
Correlate these amenity densities with available success or quality signals that Digital Atlas can provide (education metrics, income, employment in relevant sectors, population/job growth, daytime population, or other higher-order signals). Note explicitly if research-paper counts or academic rankings are unavailable.
Rank the universities by amenity density and by strength of correlation with success signals.
Produce a single, self-contained, well-styled HTML report as the final deliverable.

Required workflow

Phase 1: discover capabilities (do this first)

Call the catalog tools to list layers, families, and place categories.
Explicitly search for place categories and metrics related to: restaurant, cafe, coffee, bar, pub, sports facility, gym, stadium, arena, fitness, university, college, campus.
Note the exact category names (L1/L2) and any anchor types (e.g. "campus") that Digital Atlas actually supports.
Confirm how universities/campuses and their surrounding areas can be resolved (places, anchors, contexts, trade areas, H3, CBSA, etc.).
Identify which success-related metrics exist (education attainment, median income, workplace jobs, growth/momentum, etc.).

Phase 2: select universities

Resolve a representative set of universities (aim for 10 to 20 well-known campuses that have clear place or anchor records). Prefer geographic diversity and a mix of public/private.
For each, record name, place_id or equivalent identifier, location, and any available campus-related context.
If full national coverage of universities is impractical, document the selection criteria and sample clearly.

Phase 3: measure amenity density

For every university: define a proximity context (a drive or walk trade area of a fixed number of minutes, or a reasonable radius around the campus centroid/anchor), the same definition for all; count restaurants, cafes, coffee shops and bars, and sports facilities, gyms, stadiums, arenas and fitness centres, using the exact categories from Phase 1; compute absolute counts, density per 1,000 residents or per km² (or per daytime population if available), and any saturation, composition or accessibility metrics the atlas returns; note data-quality and coverage notes.

Phase 4: success signals and correlation

Pull available success or quality proxies for the university area (education metrics, median household income, per-capita income, relevant employment, growth rates). State explicitly whether research paper output, citation metrics, or official rankings are present in Digital Atlas; if not, use the best available proxies and label them as such. Compute simple correlations or rank-order relationships between amenity densities and the success signals within the sample, and highlight any clear patterns or lack thereof.

Phase 5: scoring and ranking

Define transparent rankings (show formulas): an amenity density ranking (restaurants/cafes/bars plus sports facilities), a success-signal ranking, and a combined or correlation-informed view. Assign qualitative labels where useful (High density / Strong signal, Moderate, Low).

Phase 6: HTML report

One complete, self-contained HTML file: clean, mobile-friendly styling; executive summary with key findings and rankings; methodology (exact tools, categories, proximity definition, releases, scoring); ranked tables; individual university sections; a data-quality and limitations note (especially research papers and missing categories); a footer with generation timestamp and "Data exclusively from Digital Atlas via MCP". No external images or scripts.

Output rules

Show your work: list the key tool calls and the exact numbers returned.
Never hallucinate a facility count, population figure, or correlation.
If sports facilities or certain success signals cannot be reliably measured, say so and score only what is measurable.
Final answer must include the full HTML report.

Begin by exploring the Digital Atlas catalog for relevant place categories (restaurants, cafes, bars, sports facilities, campuses) and available success-related metrics.

What the brief got wrong, learned in the run

"Identify universities using only Digital Atlas" needs a name list. The atlas has 160 listings in the College & University category inside the UT Austin catchment alone; there is no institution registry or enrolment field to rank by. The sixteen were chosen by name for geographic and public/private spread and then resolved with places.find filtered to the category. A future brief should name the campuses as input.
Geocoding a campus name is ambiguous. find_geography.geocode returns departments and satellite sites and asks for clarification; the category-filtered name search, with places.resolve as the fallback, is what finds the main listing.
The main listing's position decides the catchment. Columbia's listing snapped to an isolated campus path (0.55 km² reached) and was re-run from the 116th Street gate; Ohio State's listing sits at the western edge of a very large campus, so its 15-minute walk holds 14 food and drink places, against 152 from the Ohio Union side (see the sensitivity check in the report). Large campuses need a declared centre, or several origins.
No stadium, arena or campus-recreation category exists. Sports facilities are seven gym, studio and club categories; a university recreation centre counts only when it is a listed place.
No research output, citations or rankings exist, and area_profile.daytime_population returns LODES job flows (inbound, outbound, net), not a daytime estimate. Net job inflow per resident was used in its place and labelled as such.
Catchments full of students distort the resident signals. Median age near 20, median household income under $35,000 and poverty shares above 40% are what a dormitory neighbourhood looks like in the ACS; the Florida catchment covers two dormitory tracts only. The success signals describe the neighbourhood, not the university.
LODES has coverage breaks. Michigan's tract jobs fall to zero in 2023 and Stanford's jump from 4,315 (2016) to 20,454 (2021); the momentum regimes for both are artefacts and are flagged.

How the measurement was improved (measurement 2)

The first pass (190 calls, from each campus's main College & University listing) showed that the listing position decided the catchment and that the resident signals were a function of student density. Measurement 2 (149 calls, finished the next morning after two connector drops) changed three things and kept everything else identical; its tables are the report. The first pass survives in steps.md with a side-by-side comparison: the two densities correlate at rho 0.71, Ohio State, Wisconsin and Florida left the bottom once the walk started at the union, and Georgia Tech and Stanford fell.

A declared centre for every campus. The student union's street address is an input to the case, resolved by find_geography.geocode, and the 15-minute walk starts there for all sixteen. Fixing only the sparse campuses would have manufactured a correlation out of origin choices, so the rule applies to every campus.
Student density held constant. Each catchment's 18-to-24 share is recorded and every correlation is repeated as a partial correlation with that share held constant, so the question becomes whether, among catchments of similar student density, the richer or busier ones have more amenities. Momentum was dropped after LODES coverage breaks made it an artefact for two campuses.
A real success measure, outside the atlas. The atlas-only analysis is unchanged, but the outside view adds the QS World University Rankings 2026 position of each of the sixteen and correlates the atlas measures against it, labelled as outside data.

What this case shows can be done

A campus is a place. places.find with filter.categories: ["College & University"] and a state finds the main listing; its point is the origin for everything else.
One routed walk catchment is a reusable handle. routing.catchment (walk, 15 minutes) returns a cat_ handle that places.count, every area_profile action and area_change.momentum accept, so a whole neighbourhood profile is one handle and ten calls.
One count call gives the whole level-2 split. places.count on the Food & Beverage level-1 with top_n: 80 returns every level-2 count, residents, places per 1,000 residents and the price-tier mix; cafes, bars and restaurants are a grouping of that one response.
Neighbourhood signals come with their status. Attainment, income, jobs and momentum arrive with available, proxy or partial and a rule, so the report can say which numbers are true area values.
Correlation at small n is a few lines of arithmetic on the recorded numbers, and the honest result (no |rho| above 0.5 across sixteen) is as useful as a positive one.

What it suggests next

The same recipe with a declared campus centre (the student union or main gate) for every university, or several origins per campus, to remove the listing-position artefact.
Thirty to fifty campuses, which makes the correlations worth reading; the recipe costs about twelve calls per campus.
Pair the catchment with trade_area.presence and movement.visit_purposes for when people are there, which the walk catchment alone cannot say.
A drive-time variant (10 minutes) for suburban and rural campuses where a 15-minute walk is mostly campus.
What it cannot do: say anything about research output, graduation rates, rankings or student outcomes; those would have to come from an outside dataset joined on the institution.