Everyone talks about rising food costs. The national figures, the inflation panic, the pressure on household budgets. But buried in Stats NZ's food price data is a number so strange it stops you cold: Whanganui's food price measure dropped from $15,641 in 2013 to $7,798 in 2014. That's not a typo. It halved. In one year.

This isn't a cost-of-living miracle. It's a data break.

Look at the trajectory: 2010 sat at $14,652. The next three years saw steady, modest increases: $15,404 in 2011, $15,641 in 2012, holding at $15,641 in 2013. Then: freefall. The 2014 figure of $7,798 is the lowest comparable level since 1993, back when New Zealand was a fundamentally different economy.

Nobody's food bills dropped by half in 2014. Not in Whanganui, not anywhere. Which means something changed in how this was measured, collected, or categorised. Either the basket of goods being tracked got redefined, the sample size shrank, or the methodology shifted in a way that makes before-and-after comparisons meaningless.

This matters beyond Whanganui. Food price indices are supposed to track real change in what households pay. Policymakers use them. Economists cite them. Journalists write stories about them. But when a figure can halve overnight without any corresponding reality check, without any visible methodological note or caveat, it raises a bigger question: how reliable is any of this?

The trouble with official statistics is that they carry an aura of precision. We see a number, we assume someone counted carefully. But data collection changes. Survey methods evolve. Regional samples get adjusted. Sometimes a figure that looks like a trend is actually a measurement artefact, a ghost in the system.

Whanganui residents didn't suddenly start eating half as much, or paying half as much per item. Their lived experience in 2014 would have looked a lot like 2013: steady prices, maybe a bit of creep upward, the usual grind. But the official record now shows a collapse that never happened.

This is what broken data looks like. Not obviously wrong, not laughably absurd, just quietly implausible. The kind of thing that slips through because nobody's job is to ask: does this number make sense? Did anyone in Whanganui notice their grocery bills dropping by thousands of dollars?

The lesson: treat every data series with a raised eyebrow. Especially when the story it tells is too neat, too dramatic, or too divorced from what people actually experienced. Sometimes the most interesting thing a dataset reveals is that it can't be trusted.