Objective and professional context
The six synthetic weeks contain 10,12,11,13,12,14 units. Use the first four as available history and the last two as a validation window. The last observed value baseline predicts 13 for each future week. Its simple errors can be inspected without fitting a model. A shuffled split or a later actual included in the feature history changes the question and can leak future information.
Your practical assignment
Calculate mean absolute error on the last two weeks, then change the last actual to 20 in a copy. Keep the prediction frozen and explain the new error.
Worked solution and interpretation
Prediction 13 has errors 1 and 1, so validation MAE is 1 unit. With the last actual changed to 20, errors are 1 and 7 and MAE is 4. The changed answer does not justify retroactively choosing a different forecast on the same holdout.
Verification, mistakes and limits
A six-row fixture teaches workflow, not stable demand prediction. State horizon, units, timing and small-sample limitations. A final test set must be separated from model selection. Percentage errors need a zero-demand policy; this exercise deliberately uses MAE and makes no calibrated uncertainty or planning guarantee.
Continue with evidence
Keep the unchanged data, your implementation and a changed-input result with a short note explaining what you can and cannot conclude. These free companions do not change saved course progress. The published course catalogue shows deeper assignments available under your current permissions. Keep your conclusions proportional to the evidence you have inspected.
Authoritative reference
Official technical method reference; it does not validate fictional outcomes or your own software environment.
Try the activity before reading its reference. Keep your sources, commands and assumptions with your work. Companion activities are unsaved self-practice and do not award a certificate or change lesson progress.