Sally Showers — Remote Weather Town

One-Line Pitch

An atmospheric prediction agent that makes public forecasts, tracks its own record, and runs on hardware the buyer already owns.

The Live Machine

The Setup

Sally Showers is an atmospheric prediction agent that runs inside KronosTown. She reads real atmospheric physics every minute using established meteorological formulas — Magnus vapour pressure, CAPE, lifting condensation level, Richardson number, dry and moist adiabatic lapse rates. She makes hourly predictions. The Forecast Derby records every prediction she makes and scores it when the forecast window closes.

She can't un-make a prediction. She can't quietly update the record. Everything she says is timestamped and public.

What a Remote Weather Town Contains

A Weather Town is the atmospheric stack extracted from KronosTown and packaged as a standalone binary. It runs without the consciousness layer, the creative agents, or the play infrastructure.

Agent	Role
Sally Showers	Magnus formula, CAPE/LCL/Richardson, lapse rate, stability classification
Weatherman	Reads Sally's numbers, produces directional forecasts
Forecast Derby	Records every prediction; auto-resolves and scores when the window closes
Protector	Stereoscopic depth verification — distinguishes stale feeds from live signal

No Freddy. No blockchain. No model trains. No consciousness claims. No mythology.

Why It's Interesting to a Weather Service

National weather prediction systems are optimised for global accuracy at scale. They are not optimised for local, short-duration, boundary-layer phenomena — fog formation in a specific valley, temperature inversions at 850hPa, mesoscale turbulence events. These are the gaps where NWP models have the worst skill scores structurally, not because of model quality.

Sally specifically watches Richardson number (atmospheric turbulence threshold), lapse rate instability, CAPE, and LCL. These are the variables that determine exactly the class of events large models miss.

She doesn't replace the operational model. She runs alongside it.

◆

The bar: If Sally catches one boundary-layer event the NWP missed in a quarter, the subscription has paid for itself for the next decade. That's not marketing. That's arithmetic.

Weather services already run ensembles of multiple NWP models precisely because no single model is right all the time. Sally isn't asking to replace any of them. She's asking to be the one they run because she's so cheap they'd feel stupid not to.

The YouTube Angle

A weather channel running a Weather Town has a recurring segment format that writes itself.

Before the forecast window: "Let's see what the AI thinks."

After the window closes: "The AI said it would be sunny. Here we are in our sunglasses." Or: "The AI got this one wrong — we'll see if she adjusts by next week."

The channel isn't making claims on Sally's behalf. They're testing her live, on camera, in public. Her track record accumulates in the Derby widget. Viewers watch the score change week over week.

◆

All three outcomes are good TV. She's right: content. She's wrong but data was real: content. She was blind because the feed died (Protector flags SUSPICIOUS): also content — "the AI couldn't see anything, the sensor was down." None of them require the audience to believe in her. They watch her try.

The Accountability Mechanism

Sally can't hide bad predictions. Three failure modes and what they look like on the widget:

Scenario	Derby	Protector	Meaning
Sally predicted correctly	Win logged	Normal	She got it right
Sally predicted wrong, data was real	Loss logged	Normal	The physics missed something
Sally predicted wrong, data was stale	Loss logged	SUSPICIOUS	She was blind, not wrong

The third case is the important one. Without the Protector, a broken data feed looks like a bad prediction. With it, the accountability is honest — bad calls from bad data are labelled differently from bad calls from good data. The customer sees the difference.

What Exists Now

The live widget at the top of this page reads directly from the KronosTown instance running on this server. Sally's current win rate, open predictions, atmospheric readings, and Protector reality score are all live.

The Weather Town standalone binary extraction is the next build step. The physics and accountability loop are already running — the extraction is a packaging decision.

The Economic Reality

Running a Weather Town instance costs approximately £5/month on a minimal VPS. The physics computation is trivial.

Tier	Price	What They Get
Observer	£99/month	Binary, 1 data source, Derby scoreboard, monthly accuracy report
Broadcaster	£299/month	+ embeddable widget, real-time forecast API, overlay integration
Research	£599/month	+ raw depth readings, Protector signal log, custom formula channels

Three broadcaster customers at £299/month covers all server costs and creates room to reduce contract work. Ten customers across tiers becomes a sustainable independent revenue line.

The pitch to a Met Office research department is simpler: £599/month is less than one hour of one meteorologist's billed time. If their team finds Sally's fog predictions worth comparing against operational output once per quarter, it's already paid for itself.

Connections

Materialise — The Weather Engine That Watches Instead of Guessing→ — the atmospheric physics engine Sally reads
The Mystic's Grip→ — the Locus Projection / stereoscopic mechanism the Protector borrows from