I sold the Denver flash crash

I was long 95–96. PredictCast had the forecast right. The bracket settled. And I still barely made money. Informational only

Not financial advice. Personal trading journal about public Kalshi weather tape. Not affiliated with Kalshi or the NWS.

The setup

Thursday afternoon, Denver. I'm watching KXHIGHDEN-26AUG07 and PredictCast is showing a forecast high of 96°F. The 95–96 bracket (B95.5) is trading around 48–50¢ at midday — a coin flip. I buy in, thinking the forecast edge is real.

Through the afternoon, B95.5 climbs. By 4:30 PM MDT it's at 80¢. I'm up nicely. PredictCast still says 96.

Then the bottom falls out.

The crash

Starting around 5:00 PM, METAR at Denver International shows ~93–94°F. Traders watching weather apps pile into the 93–94 bracket. B95.5 collapses: 80¢ → 50¢ → 30¢ → 9¢ in seventeen minutes. By 5:16 PM my position is worth a dime.

At 5:17:24 PM, it reverses. Somebody sweeps B95.5 YES from 10¢ through 89¢ in one burst — 44 fills in a single timestamp. Within 6 seconds the price is at 97¢. Within a minute, 99¢.

But I wasn't there for the recovery. I'd already sold.

B95.5 YES price · Aug 7 afternoon MDT · Kalshi public trade data

Three paths, one position

Looking at this after the fact, there were three ways to play the same long:

Diamond hands

$1 settlement

Hold through the crash. Trust the PredictCast forecast. CLI settles at 95°F → B95.5 pays $1. Maximum payout, maximum stress.

Take profit at 80¢

~60% gain

METAR is showing slightly lower than forecast. Sell around 80¢ at 4:30 PM, lock in profit, avoid the crash entirely. The disciplined exit.

What I did
Tiny profit

            Watched it crash. Panic-sold mid–flash crash somewhere in the teens. Made almost
            nothing on a position that was right the whole time.
          

Why the crash happened (and why it reversed)

Same story as the NYC July 13 whale tape: METAR ≠ settlement.

Denver's hourly METAR was showing ~93–94°F. Weather apps echoed that. But the NWS Climatological Report — the actual settlement source — printed MAXIMUM 95°F at 3:36 PM from sub-hourly sensor data. One degree above the apps.

Between 5:00–5:16 PM, traders priced the METAR number. At 5:17 PM, someone priced the settlement number. The CLI text didn't publish until 5:29 PM — twelve minutes after the reprice.

The real lesson

PredictCast had the forecast right. The model said 96, CLI settled 95. I was in the correct bracket the entire time. But I didn't trust it when the tape went against me.

The more actionable takeaway: when METAR diverges from the forecast, that's the moment to decide. At 80¢ with METAR reading lower, the smart trade was probably to take profit and remove the risk — not to hold and hope, and definitely not to panic at the bottom.

Selling at 80¢ with a clean ~60% gain beats both diamond-handing through a flash crash and panic-selling for pennies. Sometimes the best trade is the boring one.

95°F

CLI settlement MAX

96°F

PredictCast forecast

~93–94°F

METAR / weather apps

6 sec

10¢ → 99¢ recovery

Sources: Kalshi public trades for KXHIGHDEN-26AUG07; NWS CLI product CLIDEN (BOU); PredictCast ensemble forecast. Not financial advice.