You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
Civic Forecasts is a small research project about a practical problem: public data often arrives late. I want to measure how much that changes a forecast, and make the test easy for someone else to run. The first benchmark uses weekly sanitation requests from NYC 311. I am seeking $4,800 for twelve weeks of part-time work alongside my studies at the University of Florida.
The prototype, assembled on September 6, covers 931,636 sanitation requests across five boroughs and 156 complete weeks. It compares five forecasting methods using an earlier validation period and 255 borough-week test forecasts in 2025. The selected ridge model had a mean absolute error of 138.6 requests, versus 168.8 for the previous week's count and 148.9 for a four-week average. With an extra four weeks of data withheld, the ridge model's error rose to 153.0. Coverage of its nominal 90% intervals fell from 89.0% to 87.5%.
These are retrospective results from a September 2026 data extract. The source can be revised, so the benchmark does not establish what would have been known in 2025. Requests also measure reporting behavior, not all underlying service needs.
The next phase will add a second service, archive dated snapshots, and issue forecasts before their outcomes are known. I will release the code, forecast rows, comparison methods and a short report. I will also seek feedback on one concrete planning task from potential users; there is no agency partner or agreed pilot yet. Success means a reproducible test and an honest account of whether it helps that task, including a negative finding if it does not.
The $4,800 budget is $3,000 for 120 hours of research time, $900 for independent review of the methods and data assumptions, up to $600 for computing and snapshot storage, and up to $300 for local travel and user feedback. These are estimates for future work. I will use existing computing resources first.
At the $1,500 minimum, the scope is smaller: one service, an initial sequence of dated forecasts, a methods review and a public release. That budget is $750 for 30 hours of research time, $500 for review, $150 for storage and computing, and $100 for feedback expenses. The second service and longer evaluation would wait for further support. Work would start after funding, with dates adjusted to preserve the promised evaluation period.
I am Ansh Gupta, a first-year UF student studying mathematics and statistics and working on predictive systems. This is an independent project with no paid staff. My earlier work includes a Python random-walk simulation with 300,000 trials and ongoing research on Florida detention-screening data. The latter is unfinished and has no published causal finding.
Outside research, I served as a student-government senator at Hillsborough College and volunteered with Feeding Tampa Bay. A separate nonprofit public-records project has a documented batch of 169 outreach emails covering 212 property records. Those numbers describe outreach, not funds recovered.
The largest risk is that a forecast of reported requests is a poor proxy for the decision a public-service team actually needs to make. Data revisions could also make retrospective accuracy misleading. A second service may not show the improvement seen in the first, and potential users may not find the output useful.
Dated snapshots and forecasts address the information-timing problem. Simple baselines, held-out evaluation and independent review make weak results easier to spot. User conversations will test whether there is a useful task before making any deployment claim. I have not yet demonstrated institutional demand. If that remains absent, the result will be a research and teaching resource rather than an operational tool.
No external funding has been received for Civic Forecasts. On September 6, 2026, I applied to Emergent Ventures for $4,800 for the same phase; the decision is pending. I also applied to the Foresight Fellowship's 2028 cohort, requesting compute support. Other research-grant applications are being prepared. I will disclose any overlapping award before accepting funding and reduce or decline overlapping support. This page does not represent an additional $4,800 of need if another funder covers the phase.
A separate $1,000 Awesome Foundation Gainesville application concerns a nonprofit information pilot and is also pending. It does not pay for this forecasting work.