Test the command rule
If you recognize requirements only when a sentence says "must" or "shall," you can miss work that a customer already expects. An illustrative service schedule reads: "Weekly export: CSV, delivered to the shared folder." The row names an output, a format, a frequency, and a destination. Its missing modal verb does not remove those details.
Read the surrounding agreement
A sentence-only extractor may skip that row; an aggressive extractor may turn nearby options into promises. Begin with the column labels and the schedule's relationship to the agreed service. Preserve all four fields as a qualified candidate, then determine whether the entry records an agreement or a proposal. Until evidence settles approval, keep that status unresolved. An owner, approver, or start date should also remain unknown when absent.
A hypothetical activation condition
Imagine a different, explicitly hypothetical schedule: the same export row is approved, but a heading says "During primary-feed outages only." Approval now establishes a conditional commitment, not continuous weekly delivery. The practical rule is to preserve the activation condition alongside the four fields. If the document does not show whether an outage is active, record applicability as unresolved; do not convert an approved contingency into a routine task.
Make the evaluation discriminate
Build a review set containing the weekly row, a row labeled optional, a description of the current manual process, and conventionally worded requirements. Add the hypothetical outage case to test conditional applicability separately from approval. A useful check asks whether each candidate keeps its meaning and status. Spotting "CSV" alone cannot distinguish a required delivery from an option, an existing habit, or a contingent obligation.
Keep the evidence in proportion
The project's DemaRQ research note motivates separating requirements from examples and explanation. That sentence-classification setting does not establish approval or resolve business context. These scenarios illustrate a project interpretation; they are not experimental results or a fresh reproduction of the research. ABOLTUS is in development. The justified revision is modest: command words are clues; commitments require contextual evidence, and missing facts must stay missing.