Set up a Houston-style build-out of Wing, Zipline and Amazon delivery sites using their FAA-filed flight profiles, choose the heights and rules, and run it. Your runs are computed live in your browser by the same engine used for the published studies; the two example runs shown first were computed with it when this page was built.
Lab runs use a 24 km area, 1 hour and one seed, so they are faster and noisier than the published studies (the headline uses real 68–80 km layouts, 2 measured hours and 10 seeds, with 95 % ranges). Expect the same direction and similar size of effect, not the same numbers. Try a few seeds before reading much into one run.
| Scenario | Planned conflicts | Tight | Flights delayed | Avg wait | Off-plan events | Conflicts per 100 events | Time |
|---|
Download the scorecard as a spreadsheet (CSV): one row per study with its question, result and how many registered predictions passed.
Since 22 Oct 2026 every study was written down with its predictions and fingerprinted before any code or run. Exceptions: VALLOC, VALLOC2 and ROBUST were committed together with their results, and FLEET’s analysis came after its runs. Failed predictions stay in the record.
| Study | Question | Result | Predictions |
|---|