WeatherEdge 2.0: One Plan, Fifteen Cities, an Honest Scorecard

This is a product announcement. Nothing here is financial, investment, or trading advice, and nothing WeatherEdge publishes is a recommendation to place a trade. It's meteorology, probability, and a record of how often we're wrong.

This is WeatherEdge 2.0 — the biggest change since we launched, and the first one that touches pricing, coverage and how we grade ourselves all at once.

Three things shipped together, and they're really one decision: stop making people guess what they're paying for.

At a glance

	Before	Now
Plans	Free · Pro · Max	A 7-day full trial, then Max
Markets	9 cities	15 cities
Scored on	The pre-close call	The morning call, priced

Nothing below is a migration you have to action. If you already subscribe, you keep your price and gain the rest.

One plan, and a week to decide

The old ladder was Free, Pro, and Max. Free was Houston only. Pro added four cities. Max added the rest plus the AI tools. Three tiers, three feature matrices, and a shopping decision you had to make before you'd seen the thing work.

That's gone. There is now a 7-day free trial with everything unlocked — every city, the full model, AI Insights, the coach brief, the scenario preview, and paper trading (capped at 21 trades a week — the one thing Max lifts). No credit card. Nothing to cancel; it ends by itself. After that it's Max at $24.99/month (or $229/year), and that's the whole menu.

Two consequences worth stating plainly:

The trial isn't a demo tier. It's the product. We weren't confident enough in a KHOU-only free tier to think it sold anything — you can't judge a daily-high model from one city, and asking someone to pay before they've watched it work on the market they actually trade was backwards.

If you're already a subscriber, nothing gets worse. Existing Pro subscribers keep the price they signed up at and get the full product — same bill, more markets. We're not using a plan change as cover for a price increase.

Six new cities

WeatherEdge now covers fifteen markets, up from nine:

Denver · Philadelphia · Las Vegas · Boston · Phoenix · Seattle

joining Atlanta, Austin, Chicago, Dallas, Houston, Los Angeles, Miami, New York and San Francisco.

These ran admin-only for weeks while we watched them settle against the official CLI before showing them to anyone. Two are worth flagging honestly, because you'll see it in the numbers before we do:

• Phoenix has run about a degree warm in our testing. Desert boundary-layer heating is a different problem from a sea breeze, and the model's climatology there is younger than Houston's.
• Seattle we trust least going into winter. Marine-influenced cool-season highs are the regime we have the least settled data for, and the model has a known cold lean there.

Both are live because they're useful, not because they're finished. Tuning a station's uncertainty down takes settled history none of these have yet, so all six run on the model's default — deliberately more conservative than the hand-calibrated cities. A wider σ is the model saying it's less sure, and it's the number to size off.

All fifteen are in the API too, at the same /v1/forecast call.

The scorecard change, which matters more

A subscriber wrote in to say our model "flips around all day," and that our evening emails claimed we'd "caught" trades that had actually lost. He was right about the second part, and the reason is worth explaining.

The evening recap used to score the day on the pre-close call — the model's read about an hour before market close. That number is genuinely accurate. It's also close to worthless: by then the day's peak has usually already happened and the market has repriced. Grading ourselves on it meant a day where the morning call missed — the call you could actually trade — could still render as a win.

So the pre-close track is gone from every page and email you see. It still exists internally as an instrument; it is no longer a claim we make. In its place:

• The morning call leads the accuracy page. It's the read with the widest brackets and the most room, and it's the one worth judging us on. It sits next to the NWS forecast over the identical settled days, because the honest comparison is the interesting one.
• Midday replaced pre-close as the second published number — the latest read that is still a forecast rather than a recap.
• Every "we caught it" now carries a price. Not "Caught it ↗" but "Morning missed · caught it midday at 78¢." Catching the winning bracket at 41¢ and catching it at 88¢ are not the same event, and only one of them was worth anything. Where we never recorded a price, it shows a dash — we don't reconstruct one.

Scoring is bracket containment against the settled high, not degrees. A call can land within a degree and still lose, and a scorecard that counts that as a hit is measuring the wrong thing.

The trader's read

• A free tier you can't evaluate from isn't a free tier. A week of the whole product tells you more than a month of one city.
• New markets are new model, not just new data. The wider σ on all six new cities is information, not a disclaimer — size accordingly.
• Accuracy claims are only worth the window they're measured in. Ask any weather service which hour their hit rate comes from. An accurate post-peak call is nearly free.
• A "catch" without a price isn't a result. If a scorecard won't tell you what the winning bracket cost when it was named, it isn't telling you whether the call made money.

The full track record is public — the morning call, the midday call, and the days we got it wrong. It always was. It's just measuring the right hour now.