The lasso returns a set of inputs with nonzero coefficients. Section 14.4 says the survivors are the inputs the table insists on, not the inputs that caused the outcome, and section 14.3 says ridge never deletes an input at all. The field has no procedure that guarantees the selected set is the set that matters: the list changes with the charge , with correlations between inputs, and with the particular rows you happen to hold. Two inputs carrying the same information leave the fit to choose between them, and that choice is not a claim about which of the two is doing the work.
Take a position on this question: what procedure should choose between ridge and the lasso on a table you hold, and what claim about the surviving inputs should that procedure license?
In your answer: say how you set and on which rows; say how you compare the two models; say what you report about the stability of the surviving set; say which claim you refuse to make even when the model predicts well; and name one way the procedure could mislead you when every step is followed.
Rubric. Tick each criterion your answer meets, then save the self-assessment. The platform does not grade it.
Worked solution
There is no single correct answer, so this is a model of the shape a complete answer takes, not a key. A level-3 answer begins by fixing the rows that choose the charge and the rows that report the result, or by naming a source of rows outside the table. On the first part it sweeps for both penalties and records predictive performance on rows the fit never saw. It compares ridge and the lasso at their best charges on the same held-out rows, and it reports the comparison as an estimate with its uncertainty rather than as a verdict.
The answer then says what it will claim about the survivors. It reports the set at the chosen charge, and it repeats the fit on resampled rows to show which inputs survive every time and which appear only under some draws. An input that survives every resample is a stronger candidate than one that enters and leaves, but neither is a cause. The answer says plainly that the lasso reports which inputs the fitted model uses, not which inputs produce the outcome; a correlated pair may swap places while the predictions stay the same, and an input that does nothing can survive because it happens to track the rows.
The answer names the cost: the held-out rows are removed from the fit, so the final model is fitted on less data than the table contains, and a small reporting set makes the comparison between ridge and the lasso noisy. It names a failure: if the held-out rows differ from the population the reader cares about, the chosen charge and the survivor list can both be wrong while every step is followed. A strong answer says what would change its mind, such as a survivor that keeps its place under resampling and still improves prediction on new rows, and it states the population assumption that the whole procedure rests on. An answer that scores 0 says only that the lasso selects and ridge shrinks, and stops there.