Creeper

New and used are evaluated separately. We take the last estimate each week, then compare a recent price trend with keeping the last price unchanged.

Each check predicts an observation 23–37 days later using only earlier data. At least four separate checks are required. A trend must reduce average error by at least 10%; otherwise we hold the last price.

We withhold the outlook if prices are stale, recent changes are unusually large, or average percentage error exceeds 20%. These are screening rules, not a confidence rating. The same checks select the model and report its past error, so future accuracy may be worse.

No assumed price boosts for retirement, popularity, or rarity. This projects our market estimate, not the price a specific listing will sell for.