Prompting only： How Much of Philadelphia Did I Actually Use?

Day 10 (Prompting only): Today’s map was made entirely in a chat with Claude Code on the command line, from my TripLibre travel diary. I wrote no code and picked no design. I only wrote prompts and judged the results. Claude chose the idea, wrote the R script, rendered the map and wrote the caption. The finished map sorts all 159 Philadelphia neighborhoods into three plain groups for Sept. 28 – Oct. 30, 2026: places I stopped (shaded by hours), places I only passed through, and places I never went. I stopped in only 13, about 14% of the city’s area. I passed through 62 more and never set foot in the other 84. University City alone took 75% of my hours out (98 hours).

The prompts, in order (prompts 4 and 5 were originally in Chinese):

1. “Today’s map completely up to you, use my TripLibre data (make a map you want to make), only requirement, different than before.”Claude picked a 3D space-time aquarium, Hägerstrand-style. The city lay on the floor, height was the time of day, stops were vertical pillars and trips slanted between them.
2. “No one would understand this map, could you make it easier to understand, probably flatten out some lines.”Claude flattened it into a map strip above a train-timetable chart sharing one west–east axis, with one day narrated step by step.
3. “It’s still too complicated, try another way.”Claude made a plain dot map of places colored by morning, afternoon or evening: “Days at Penn, nights on Broad Street.”
4. “This is pretty much the same as before. Can you make it a bit different?”Claude switched the form itself to a neighborhood choropleth, a form none of my earlier days used, around one question: how much of the city did I actually use?
5. “Fix the inconsistent internal boundaries on the map.”Claude found about 250 sliver gaps and 600 tiny overlaps in the neighborhood polygons, cleaned them into a proper coverage and redrew every inner border once.
6. “Write an MD in the mapchallenge folder, following the earlier style, but emphasize the prompts we used, in English.”This write-up.

What I learned from prompting only: the first answer was the most ambitious and the least readable. Each round, I gave feedback as a reader (“no one would understand this”, “still too complicated”, “pretty much the same as before”) rather than as a designer. That kind of feedback pushed the map toward one simple question and away from a clever form. Claude kept the facts honest the whole way: every number on the map is computed from the diary, and it fixed its own bugs along the way, such as a share that first came out at 100% and then 0% before it was right.

Technical Implementation:

• AI: Claude Code (Claude Opus 5.5) wrote, ran and visually checked every version of the script from the prompts above. I reviewed the outputs and the privacy rules.
• R with sf, dplyr and ggplot2 for classifying neighborhoods and mapping
• ggtext for rich-text annotations and showtext with Bricolage Grotesque and IBM Plex Mono for typography
• Method: a neighborhood is stopped if it holds at least one non-home stop, shaded by total hours there. It is passed through if any recorded trip route crosses it without a stop, and never otherwise. Neighborhood polygons were cleaned into a gap-free, overlap-free coverage before drawing.
• Data: personal TripLibre travel diary export (visits, places & trip routes). The raw data stays private. Home is not counted, nor is any stop within 1 km of it, and results are shown only at neighborhood level.
• Basemap: Philadelphia neighborhoods from Azavea via OpenDataPhilly; water from OpenStreetMap contributors