The asteroid that was a certainty on the day NASA struck it off the list
Showing the misleading chart
Thirty-two of NASA’s own updates on the chance that asteroid 2024 YR4 hits Earth in 2032 climbed from 1 in 1,040 to 1 in 32 in seven weeks. Fit an exponential through them, carry it forward, and a planning memo we drew for an invented wedding venue concludes the impact is a certainty by 2 April 2025. On 2 April 2025 NASA removed the impact from its risk list. The odds had risen because the range of possible paths was shrinking around Earth, and five days after the memo they were 1 in 26,000.
01The claim
NASA’s chance that asteroid 2024 YR4 hits Earth in 2032 has doubled every 9.4 days for seven weeks (R² = 0.945), so at this rate the impact is a certainty by 2 April 2025.
02The trick
Every probability on the memo is NASA’s, and the exponential fit through the 32 days with an archived solution from 30 December 2024 to 18 February 2025 is computed correctly. The trouble starts at the last point, where the line carries on. A fitted curve can only describe the points it was fitted to. Past 18 February it is a claim about a quantity whose possible endings are known in advance, and that quantity is not a growth rate. An impact probability measures how uncertain the asteroid’s path still is, and for a single pass it has two possible endings: 0% if the asteroid misses and 100% if it hits. More observations push it towards one of them. And if the forecaster is doing its job, today’s number already prices in where tomorrow’s is expected to go: at 3.1%, the honest expectation was a 3.1% chance of ending at 100% and a 96.9% chance of ending at 0, not a climb. While Earth sat inside the range of possible paths, each observation that narrowed the range made Earth a larger share of what was left. JPL’s 3-sigma range on the 2032 miss distance narrowed from ±9.55 million km on 6 January to ±458,000 km on 18 February, about 21-fold, and the odds rose about 22-fold, from 0.14% to 3.1%, roughly in step. That climb was the sharpening itself, so its slope says nothing about where it ends. ESA calls it a well-understood pattern: the odds often rise at first, then drop to zero as the region of possible orbits shrinks and moves away from Earth. The memo’s own choices are knobs as well. Fit only the Torino-3 weeks, or use a straight line instead of an exponential, and the same points reach certainty anywhere from 26 May 2025 to 21 December 2029. All four recipes reach certainty before the asteroid arrives, and the outcome was 0%. (The memo is our own demonstration, dated as if written on 18 February 2025. The venue, its committee and the recommendation are invented, and every probability is NASA’s.)
03The fix
The redraw stops the gold curve at 18 February, marks the 32 fitted updates as a shaded band, and shows the extension as what it is: a red dashed line that passes 4.8% on 20 February and 6.0% on 23 February. Then it plots what Sentry actually published next. The odds were 1.5% the following day and 0.27% the day after. By 23 February they were 0.004%, 1 in 26,000, and Torino 0, and by 18 March they were 1 in 5.3 million. On 2 April 2025, the day the memo’s curve reached 100%, NASA removed the 2032 impact from its risk list. A second panel draws the uncertainty the probability summarises: JPL’s nominal miss distance and its 3-sigma range, against Earth at zero. Every range up to 22 February reaches Earth, and from 23 February none does. It is a rough proxy, not Sentry’s own geometry, but it shows the mechanism: while the range straddled Earth, each narrowing raised the odds; once the best-fit path drifted away and the range stopped reaching Earth, they collapsed. The memo had a warning in its own window, too: on 18 February its curve sat at 4.2% against an observed 3.1%, and every update from 10 February on had come in below it. A third panel runs four recipes on the same points, two shapes and two windows, and their certainty dates spread from 2025 to 2029. When a series can fan out that far, no single line through it means much. The Moon shows the same pattern more slowly. Its odds rose from 0.8% in February 2025 to 4.3% after Webb looked in May, then fell to zero in March 2026. The general fix is to stop the ink at the last observation and ask what the quantity is allowed to do. A share, a count or a probability has bounds and often a known mechanism. When one exists, as JPL’s orbit model does here, use the forecaster’s mechanism rather than a slope through its outputs.