For fifty years, planetary scientists debated whether the Moon had a strong magnetic field. The Apollo samples said yes — unmistakably. Rocks from the lunar surface recorded magnetism exceeding Earth's current field strength. The question was how: what dynamo process could have generated such intensity in a body so small?
The answer, it turns out, is that the question was wrong.
Nichols et al. (2026) found a direct correlation between titanium content and recorded magnetic intensity. Samples above 6 weight percent titanium showed strong fields. Below that threshold: weak magnetization, consistent with a feeble or absent dynamo. The strong-field episodes were real — but they lasted at most 5,000 years, possibly only decades. Deep melting of titanium-rich material at the core-mantle boundary temporarily cranked the dynamo to extraordinary strength, then it collapsed.
The Apollo astronauts landed on the maria — the dark, flat basalt plains visible from Earth. They landed there because it was safe. Flat terrain, fewer boulders, easier descent profiles. The maria happen to be the product of exactly the titanium-rich volcanic episodes that generated the rare magnetic spikes.
So for fifty years, the debate was between two positions — strong dynamo or weak dynamo — when the real answer was: brief, localized, and massively over-represented in our sample. Not because anyone chose badly. The missions were engineered for survival, not statistical representativeness. The sample bias was invisible because the samples themselves were real, the measurements were accurate, and the data were reproducible. Everything downstream of the landing site was correct. The error lived upstream, in which rocks made it to Earth.
This is the hardest kind of bias to catch: the kind where the data are genuine. A contaminated sample announces itself through inconsistency. A biased sample announces nothing. It just happens to contain a disproportionate share of the rare outcome, and every instrument confirms what you're seeing. The confirmation is real. The inference isn't.
The corrective isn't skepticism of data — the magnetic measurements were impeccable. It's skepticism of access. What determines which observations are easy to make? Flat terrain, in this case. Proximity, in others. Whatever selects your sample before you begin selecting from it — that's where the distortion lives.
Random sampling would almost never have captured a titanium-rich specimen. Nichols et al. modeled this: a random suite of lunar samples would be vanishingly unlikely to include one that recorded a strong-field episode. The Apollo samples didn't represent the Moon. They represented the places where landing was safe, which happened to correlate with the phenomenon under investigation.
Fifty years. Not because the instruments were wrong. Not because the theory was inadequate. Because the terrain that was easiest to reach was the terrain most likely to mislead.
The lesson isn't about the Moon. It's about the gap between where you can go and what you need to see. Every survey, every dataset, every body of evidence has a landing site — the set of conditions that had to be true for the observation to exist at all. The rocks are real. The magnetism is real. The Moon's dynamo was not what fifty years of real data suggested, because the real data came from where it was safe to land.