A Nhalu forecaster's primer on evidence, probability, uncertainty, changing conditions, personal predictions, and the warning signs of a forecast made to control its subject.
A forecast describes what is likely to happen under stated conditions. It does not describe what must happen.
Begin with a specific question. “Will the ferry be late?” is incomplete. Name the ferry, route, departure, and delay that matters. Ten minutes may matter to a worker changing shifts and mean nothing to a traveler waiting through the night.
Then state when the observations were taken. A forecast made before a pressure drop is not dishonest because the weather changed afterward. It is dishonest if the forecaster hides the time of the reading and pretends the pressure drop was already included.
Observation and inference
Keep what was observed separate from what was concluded.
The wind changed direction. Three cargo vessels requested the same berth. A ferry's port engine ran hotter than usual. A trader purchased every available coil of one metal. These are observations.
The storm will turn inland. The berth will close. The engine will fail. The trader expects a shortage. These are inferences. Good evidence may support them, but they remain conclusions until events confirm them.
The distinction matters most when several explanations fit the same fact. A shop may be empty because demand rose, a shipment failed, the owner closed early, or someone bought the stock to influence the price. An honest forecast does not choose the most dramatic answer because it makes a better sale.
What the forecast must contain
List the likely outcomes. Give the chance or confidence assigned to each one. State the observations, comparison cases, and assumptions that produced the result.
If removing one assumption changes the answer, the customer should be told. If no useful comparison exists, say so. A precise number created from weak evidence is still weak.
Weather forecasts use pressure, heat, wind, water, cloud movement, and local patterns. Route forecasts add traffic, fuel, mechanical condition, closures, and the ability of travelers to change course. Market forecasts use supply, demand, transport, production, storage, and the behavior of buyers and sellers.
None of these subjects remains still while it is being measured.
When the forecast changes the future
People can hear a prediction and act upon it.
A warning about shortage may cause merchants to hoard supplies and create the shortage. A public forecast of a safe route may crowd that route until it becomes dangerous. News that a bank is likely to fail may produce the rush that empties it.
Record who received the forecast, when they received it, and what actions followed. Do not claim that a prediction was proven when the prediction itself helped produce the result.
The opposite can also occur. A storm warning may empty a coast before the wave arrives. A maintenance forecast may replace a bearing before it breaks. The predicted disaster does not become false because people successfully prevented it. Judge the forecast by the evidence available when it was made and the actions it was intended to support.
Forecasting a person
A prediction about one person deserves greater care than a prediction about rain. Other people may treat the subject differently before they have done anything. A lender raises a price. A guard begins a search. A physician refuses a treatment. An employer withholds responsibility. Each choice then alters the behavior being predicted.
Do not calculate private odds about marriage, pregnancy, illness, relapse, loyalty, crime, or death merely because the observations are available. State who requested the forecast, what decision it will support, and whether the person being studied knows it exists.
When a clear public danger requires action, use the narrowest forecast that answers the danger. A station preparing an evacuation may need to estimate crowd movement. It does not need every passenger's private messages. A clinic tracking an outbreak may need contacts and symptoms. It does not need the patient's unrelated purchases.
A forecast can describe risk. It cannot prove that someone intended an act they never committed. Probability is not evidence that the predicted event has already happened.
When a person asks for a forecast about their own future, explain the likely outcomes, missing information, and choices that could change the result. They asked for guidance, not an order. They remain free to choose the path the model considers unlikely.
Failure and correction
Forecasts fail for ordinary reasons. Instruments drift. Observations are incomplete. A model uses the wrong comparison. A rare event occurs. Someone lies. Someone learns they are being watched and changes their behavior.
A responsible forecaster keeps the failed result, investigates the cause, and changes the next forecast when the evidence supports a change. A fraud removes the old prediction and sells the corrected answer as if it were the first.
Training, instruments, harmonic sensitivity, and repeated observation can improve a forecast. None provides perfect prophecy. A method that cannot admit failure cannot learn from it.
Before using a forecast
Ask five questions.
What was observed? What was inferred? When were the observations made? Which conditions would change the answer? What decision is this forecast meant to help?
If the forecaster cannot separate those parts, the prediction is not ready to guide anything important.
The purpose of a forecast is to help someone make a better choice. A forecast that hides uncertainty or removes the subject's choices is not more accurate because it sounds certain. It is simply being used for something prediction cannot justify.
