A standard learning-curve pattern for high avoidable bias occurs when, at the largest training-set size, training error remains far above the desired performance level. If the gap between training error and dev error is small, this indicates small variance despite the large avoidable bias.