Update your forecast incrementally as new evidence arrives

Treat your forecast as a probability that should shift with each new piece of evidence, not a commitment that survives contradiction.

Why it works

Forecasts that are not updated in response to evidence are not forecasts — they are commitments in disguise. The updating mechanism is Bayesian: new evidence shifts the probability, with the magnitude of the shift proportional to the evidence’s diagnostic value. The failure mode is updating too slowly (anchoring on the original estimate) or updating too violently (overweighting recent salient events).

How to do it

1. Set a scheduled check-in cadence for significant forecasts — weekly for short-horizon, monthly for long-horizon.
2. At each check-in, ask: "What evidence arrived since last time, and in which direction does it push?"
3. Make a numerical update (even if small) so the record shows a decision to hold or move.
4. Track your updating patterns over time — systematic under-updating or over-updating are diagnostic.

Evidence

Tetlock’s forecasting tournament research found that superforecasters update their predictions more frequently and in smaller increments than average forecasters, and maintain calibration over time as a result. Bayesian updating is the formal framework underlying this behavior. (observational)

In practice, people update more readily when evidence confirms their existing view. Disciplined Bayesian updating requires resisting this confirmation asymmetry, which is a sustained behavioral commitment, not a one-time technique.

Sources

• Tetlock & Gardner (2015), Superforecasting
• Tetlock (2005), Expert Political Judgment

Common mistake

Updating the narrative ("the situation has changed") but not the probability estimate, so there is no legible record of whether the updating was appropriately calibrated.