§03The forecast
Can we predict where they go?
Short version — yes, about 49% of the time. Out of 200landed flights we've scored, our guess named the right destination on 97; 103 we got wrong. Same-metro hits (e.g. picked TEB, landed JFK) count as correct because the destination city was right.
How it works — Every active flight gets a fresh hourly call. The model sees the owner's briefing (companies, residences, recurring routes, controversies), their last 20 flights, and the plane's current position + heading + speed. It returns a destination ICAO, one or two alternates, and a confidence score. When the plane lands we compare to the real arrival airport. No web search per call — the call is offline and reproducible.
This week's picks
Pre-departure predictions made before the wheels are up. The desk picks ~12 names a week based on public events, tour schedules, and corporate calendars. Each pick gets graded when the matching flight lands.
- PassedKylie Jenner→ KEWR · New York TeterboroBy May 4· Window closedExpected attendee at the 2026 Met Gala on May 4 in New York City based on leaked guest list and history of attendancePicked May 4 · 85% conf
- MissElon Musk→ KVBG · Vandenberg Space Force BaseBy May 4SpaceX Starlink mission launch scheduled for May 5 from Vandenberg, California, which Elon Musk often attendsPicked May 4 · 75% conf
- MissKim Kardashian→ KEWR · New York TeterboroBy May 4Confirmed attendee at the 2026 Met Gala on May 4 in New York City (source: Page Six guest list announcement [web:237])Picked May 4 · 95% conf
In flight now
3 liveLive AI predictions for every tracked plane currently in the air. Each card lists the most likely destination, alternates, and the desk's reasoning. Cards disappear when the plane lands — see “How we did” below for the score.
The aircraft is heading northeast at 48° from Utah, and all ALIGNED candidates (Δhdg ≤ 25°) are European airports. The Emir's recurring destinations include LFPB and EGLL, and recent flights show multiple Paris-area arrivals (LFPG, LFPO, LFPB). Past predictions incorrectly guessed KADW (off-track), but geometry now strongly supports a European destination. LFPB is both a recurring destination and directly aligned (Δhdg 10°), making it the top pick.
No ALIGNED candidate (Δhdg ≤25°) in the table. The closest two are HEGN (Δhdg 46°) and HECA (53°). HEGN was the most recent flight destination (OMAA→HEGN on 2026-06-16), and the current heading 241° roughly points toward Egypt, requiring a moderate turn. HECA is a recurring destination and mentioned in the briefing (Cairo). Confidence is low because the plane is not aligned with any listed airport and past predictions for this person have been 0/3 correct.
The aircraft is heading 70° over the North Atlantic at 594 kt, and all candidate airports have Δhdg > 90°, meaning none are aligned with the current track. This suggests the plane is still in an early climb or turn phase, likely toward a destination in the eastern US. KMDW is the closest recurring destination (1592 nm, Δhdg 155°) and appears in recent flights (KPHL→KMDW on 2026-06-17), making it the most plausible final destination despite the large heading offset. Confidence is moderate because the geometry is ambiguous and no ALIGNED candidate exists.
How we did
1717 correct2434 wrongEvery prediction we've scored, newest first. Each row shows what we picked, where the plane actually landed, and the verdict. Green = correct destination (same airport, an offered alternate, or a same-metro field). Red = different city.
How the desk works
+ Open
- 1Each owner has a briefing. Companies controlled, primary residences, recurring routes, controversies. 100–150 words, distilled by hand from public OSINT (FAA registry, SEC filings, news, financial press, Wikipedia). Nothing private, nothing scraped.
- 2For every active flight, the model sees: the briefing, the plane's last 20 completed flights as origin→destination pairs, and the current position (lat/lon, altitude, ground speed, heading). No web search, no real-time news.
- 3The output is structured: one destination ICAO, one or two alternates, a confidence score, an ETA, and a 3–4 sentence reasoning. When the plane lands we resolve its arrival airport via the same ICAO index and score the prediction. Same-metro arrivals (e.g. picked TEB, landed JFK) count as correct because the destination city was right.
We don't use real-time web context for the per-flight call — it's offline and reproducible. The weekly "This week's picks" section above DOES use web context (one Grok :online call per week), but the per-flight loop is fully data-driven from the briefing + history + position.
Score, last 30 days
Daily verdict roll-up. Bars are stacked: correct, wrong. Reads left-to-right, oldest to today.