A planning packet is long, and a short token limit quietly drops the rezoning clause the label depends on. What factor belongs in the experiment, not in hidden preprocessing?
Select an answer to reveal the explanation.
Short Explanation
A short token limit can drop the rezoning clause the label depends on. Treat max sequence length and overflow handling as a scored factor, not a silent default. Tidying every packet to one sentence, or scoring leftover fluency, hides that cutoff.
Full Explanation
Maximum sequence length and how overflow is handled change what the model actually sees. If a planning packet is truncated before the rezoning clause, the label can look unlearnable even when the architecture is fine. That cutoff is therefore an experimental factor that must be scored, not a silent default. Fluency of the leftover text is not a substitute for the task metric.