With no network this cannot forecast. What it does is make you your own forecaster, from the barometer, the sky and the wind. Celestial is what you do when GPS dies; this is what you do when the sat comms die.
Your longitude
Needed for local solar time, so the daily atmospheric tide can be taken off the pressure trend before anything is said about it. Without it, an ordinary tropical afternoon reads as a system approaching.
This reasons only from what you logged. It is not a forecast: no model, no chart, no GRIB. It is you, reading your own barometer.
What the log says
Show it as the terminal does
Yours first
Nothing of mine until yours is written down. If the app prints its guess before asking for yours, you have not forecast anything — you have read an answer and agreed with it. It feels like learning and is not.
Forecasts are made for a synoptic hour, so the observation that verifies one is an entry somebody would have made anyway.
Your call
Leave a field empty to leave it out. A forecast is several claims and can be two thirds right, so each is scored on its own.
Now mine
Log the observation this was made for when it comes, then look at the score. Three are scored, not one: scoring only you would make this an oracle, and it is not one.
The score
Weighted error points, from the WxChallenge: half a point per knot, one per millibar, a tenth per degree of wind direction — so a near miss scores like a near miss. Lower is better.
Why persistence is on the sheet
“In twelve hours it will be much as it is now” is the honest floor. It is hard to beat at short range, and a forecast that does not beat it has shown no skill at all. Standard practice in operational meteorology, and free to compute.
Scoring the rules as well as you keeps the app honest — its misses go on the record — and teaches calibration: a rule right seven times in ten is useful once you know it is seven in ten. And it lets you eventually win, which is the point. A training tool that cannot be outgrown is badly built.
Show it as the terminal does
From a 500 millibar chart
This app cannot fetch a chart and never will. But a 500 mb chart arrives by HF radiofax, on an SSB receiver, at sea, with no connectivity at all — which is exactly the situation the rest of this is built for. Read three numbers off the fax and the rules do the rest, showing every step.
Show it as the terminal does
Lessons
Each one earns its place by changing what the app says or what you would do. One of them says plainly where that stops being true, which is the point of the rule.
The honest limit
With no network this cannot forecast. No GRIBs, no synoptic charts, no models. What it does is make you your own forecaster from the barometer, the sky and the wind.
It works entirely in the observation column of the three kinds of weather data — ground truth you measured. It cannot forecast, but what it reasons from is the one category that is never wrong, and the one still available when the antenna comes down.
One shared log
This reads the observations deck-log keeps, on this device, in the same browser storage — one shared store, as it should be. The watch stander says the report once.
Never tell before you ask
The forecast screen shows nothing of the app’s reasoning until your own call is committed to the log. Not greyed out, not collapsed, not behind a tap — absent. House rule, and it came from the colregs lights quiz telling you the aspect and then asking for it.
Where the rules come from
The 500 mb rules are from Sienkiewicz and Chesneau, Mariner’s Guide to the 500-Millibar Chart, NOAA — public domain. The scoring is the WxChallenge’s weighted error points. The observation follows the VOS standard. Sources are in docs/SOURCES.md; nothing in the lessons is copied from anywhere.
Offline
Checking…
Rules of thumb: good ones, and still rules of thumb. What is tested is that the app applies its stated rules correctly and reports its own score honestly — not that the forecast is right. It will often be wrong; that is the nature of the subject, and it is why both scorers exist.