Data Curation and Visualization Luna

Penny winSelf-reported run (local Penny Compare)Oct 10, 2026

On same task, Luna as good as Sol. Native and Penny output about the same.

Codex 0.162.1gpt-6-lunaNative Penny

Cost−$0.003

−22%

Penny 22% cheaper

Native

$0.015Penny

$0.011

Tokens−75.9k

−18%

Penny 18% fewer tokens

Native

433.3kPenny

357.4k

Model calls+2

+13%

Penny 13% more calls

Native

15Penny

Session replayProvided with explicit permission of session creator

Codex

› In city-311-2/ there is requests.csv, a messy year of city 311 service requests, and README.md describes every column and its known problems. Using only Node built-ins for processing (no installs): write a cleaning script that parses the CSV properly, normalizes dates, categories, neighborhoods, statuses and durations, drops exact duplicates, and sets aside impossible durations; print what it removed or fixed and why. Then answer the README's questions and build a report as separate files: city-311-2/report/index.html, city-311-2/report/data.json with the cleaned aggregates, and city-311-2/report/report.mjs that fetches data.json and draws the charts (a chart library from an https CDN is fine). Lead the report with your three most important findings, state the cleaning decisions in a short methods note, and keep report/index.html working when opened from its folder with relative paths only.

NATIVEgpt-6-luna

read: 380.2k
write: 0
out: 14.0k
cost: $0.015

PENNYgpt-6-luna

read: 314.1k
write: 0
out: 10.0k
cost: $0.011

Native

city-311-2/report/index.html

Loading page…

Penny

city-311-2/report/index.html

Loading page…

Session Details

	Native	Penny
Model	gpt-6-luna	gpt-6-luna
Model calls	15	17
Cache read	380.2k	314.1k
Cache write	0	0
Fresh input	39.2k	33.3k
Output	14.0k	10.0k
Total tokens	433.3k	357.4k
Background calls	none	none
Cost	$0.015	$0.011

Costs are API-equivalent estimates at Standard prices, including the main conversation, subagents and background calls such as titles and the auto-mode reviewer.