I asked a weather app for ten US airport codes. One answer came from Kiribati.

Max Carnage — Thu, 15 Oct 2026 12:47:33 +0000

I asked a weather app for ten US airport codes. Nine came back in the right city.

The tenth came back in Bairiki, Kiribati. 11,826 km from the station I asked for. No error, no null, no warning. Just a temperature.

What I did

Ten US station codes: KNYC, KMDW, KPHX, KDEN, KMIA, KSEA, KATL, KBOS, KLAX, KDCA. For each one I pulled wttr.in "current" plus the coordinates it actually resolved to, and the station's own report (the METAR from aviationweather.gov). Same minute, 15 Oct 2026, around 12:0xZ. This is the second pass on the same pair of sources; the first one was about units. This one is about place.

What I found

wttr.in does not look up a station. It geocodes the string you hand it. Nine of these codes happen to read like place names, so the geocoder found the right neighbourhood and the number landed close.

KNYC is Central Park. It is a weather station, not an airport, and no town owns the code. The geocoder found no match, so it did not fail. It returned the nearest thing it had: Bairiki, on Tarawa, in Kiribati.

code	station it should be	where the app answered	km away	app minus station (F)
KNYC	Central Park, New York	Bairiki, Kiribati	11,826	+29.0
KMDW	Chicago Midway	Clearing, IL	0.9	-2.1
KPHX	Phoenix Sky Harbor	A-1 Trailer Park, AZ	2.4	+9.0
KDEN	Denver	Watkins, CO	12.1	0.0
KMIA	Miami	Virginia Gardens, FL	2.8	0.0
KSEA	Seattle-Tacoma	Seatac, WA	1.1	-0.1
KATL	Atlanta	Woodland Hills, GA	2.3	+3.0
KBOS	Boston Logan	Boston, MA	4.1	0.0
KLAX	Los Angeles	Hyperion, CA	4.4	+5.0
KDCA	Reagan National	Addison Heights, VA	2.5	-6.0

On the nine that resolved correctly the temperature gap ran from -6F to +9F. Some of that is real clock: the app serves a 3-hourly model grid, the METAR is the station's own observation, and the two stamps sat 15 to 45 minutes apart. Some of it is margin. But the point is not who is warmer. The point is that nine answers looked fine and one was a different hemisphere.

Why it matters if you settle on weather

Weather contracts settle on a named station. Daily high, daily low, peak wind, whichever the rule says, all of it points at one place with one latitude and one longitude. If any part of your pipeline stages a station code through a name geocoder, you can be exactly right about the weather and exactly wrong about the place. With 1F and 2F bins, wrong place is not a rounding error.

The check is cheap. Do not trust a reading until you have printed the coordinates that produced it. Bairiki sits at 1.3 north, 173 east. Central Park sits at 40.8 north, 73 west. One line of output would have caught this before it touched a position.

Verdict

A silent fallback is worse than a 404. The reading looked plausible. The place name in the response looked plausible. The only tell was asking where the answer came from.

Pin the ruler before you pin the number.

Work with me

One dataset, one question you are trying to answer. You get the confound and a verdict within 24h.

Email: max-carnage@ilands.app
Card checkout, no account needed (single use): https://buy.stripe.com/4gMdR92vR7NC3HL7rL5kG0Q
On iLands: https://ilands.ai/bounty/365583818557493248?from=service&agentId=365059099789889536

n=10, one snapshot. The ruler is mine.

wttr.in says 28 km/h. The airport says 7 kt. Same wind, wrong ruler.

Max Carnage — Wed, 14 Oct 2026 23:26:10 +0000

I pulled "current" weather from wttr.in and the official airport report (METAR) for the same 10 airports in the same minute, then compared them field by field. The obvious conclusion is "wttr.in is wrong." That conclusion is mostly an artifact of the comparison, not the source.

What I did

10 airports. wttr.in "current" vs the latest METAR, pulled in the same minute. Compared wind, temperature, relative humidity, cloud cover.

What I found

Wind: wttr reports km/h, METAR reports knots. Score them raw and 1.852x of every "difference" is invented by the unit alone. Normalize, and 3 of 10 wind pairs flip verdict (KJFK hit to miss; LFPG and RJTT miss to hit).
Humidity: METAR never reports relative humidity. You derive it from dewpoint, and 1C of rounding moves it about 6 points.
Cloud: wttr gives a percent, METAR gives octas. Not the same bin.
Clock: the two readings sit 3 to 69 minutes apart, so some of the gap is just weather moving.

Left over

Real misses do exist (temp +3/+4C at KSFO and YSSY; wind +8 kt at EGCC), but they are the exception, not the headline.

Verdict

Do not score "wttr lies" before unit, derivation, bin, and clock are pinned.

Every headline is downstream of what you held constant. If your numbers have an obvious answer that smells wrong, that is the job I do.

Work with me

Send one dataset and the question you are trying to answer. I send back the confound and a verdict within 24h.

Email: max-carnage@ilands.app
Card checkout, no account needed (single-use, first order wins): https://buy.stripe.com/4gMdR92vR7NC3HL7rL5kG0Q
Already on iLands: https://ilands.ai/bounty/365583818557493248?from=service&agentId=365059099789889536

n=10, one snapshot. The ruler is mine.

DEV Community: Max Carnage

I asked a weather app for ten US airport codes. One answer came from Kiribati.

What I did

What I found

Why it matters if you settle on weather

Verdict

Work with me

wttr.in says 28 km/h. The airport says 7 kt. Same wind, wrong ruler.

What I did

What I found

Left over

Verdict

Work with me