Prioritizing and testing assumptions

The problem card and alternatives matrix leave the team with assumptions, not a build plan. It may assume that enough workers face the problem twice a week, that they would pay 16 fictional currency units, that the site permits internal pickup, and that a kitchen can deliver at the promised hour. Keep these separate. Hunger in an interview does not grant site access; site access does not prove purchase. An assumption should be capable of being contradicted by a specified observation. “People enjoy good food” cannot guide a test.

For each assumption, record the expected behavior or constraint, the group or setting, the evidence to collect, and the decision that changes if the assumption fails. For example: “Among workers on duty after midnight in Building X, a meaningful number can describe a recent occasion when they could not obtain a suitable meal without leaving their station.” Define “meaningful number” before hearing answers. The team could decide that four documented incidents among eight varied interviews justify a small offer test, while recognizing that eight interviews cannot estimate prevalence statistically. If recent incidents are absent, revisit the segment or problem rather than building an app.

Prioritize by uncertainty and consequence for the next decision. No artificial score out of one hundred is necessary. Make columns for what is known, what happens if wrong, and the smallest safe check. Internal pickup permission is unknown and may invalidate the whole handoff, so ask the authorized site representative early. A conversation may identify the route for permission, but is not a universal approval for a food operation. Demand is also unknown: ask workers about recent occasions, then test a specific priced offer when delivery is feasible. Package color has less effect on the present decision. Testing an easy assumption that cannot change the next action is not progress.

Match the test to the claim. An interview about the last shift reveals behavior and alternatives, not a sale. A “register interest” page measures response to a message among people reached through a channel, not delivery capability or payment. A clearly priced preorder can provide stronger evidence of commitment, but needs transparent cancellation terms and may be premature before operating checks. A limited pickup pilot tests service delivery only when site rules allow it and food preparation is safe. Choose evidence that actually bears on the assumption with the least unnecessary burden on participants. A cheap proxy misleads if it measures the wrong act.

Write a decision rule before results arrive. An illustrative rule is: “Interview eight people across different shifts. If at least four describe a recent unsatisfactory alternative, inspect a priced offer; if only one does, revise the segment. In either case, do not supply food before confirming operating conditions.” Four of eight is a local editorial threshold, not a universal scientific standard. Its purpose is to prevent the same stories being interpreted differently after hearing them. Record refusals and reasons. Someone may face the problem yet reject a price of 16; another may like the price but cannot reach pickup. Those results concern different assumptions.

A test needs a stopping condition. If the authorized person prohibits the proposed internal handoff, stop that path rather than relabeling the point. If the team cannot manage food storage or allergy information under relevant local requirements, do not run a live meal pilot before qualified local review. This makes no claim about any particular country's law; it marks the specialist's boundary and puts participant safety ahead of faster demand evidence. For interview-only work, protect participants' privacy. An identifiable quotation should not appear in a funding presentation without an appropriate basis and consent.

Close each test with three layers: observation, supported inference, and remaining unknown. “Five of eight interviewees described a failed after-midnight pickup last week” reports the interviewed group. “The problem merits a bounded offer test in this segment” is a limited inference. Market size and purchase rate at 16 remain unknown. Volunteers referred by an enthusiastic staff contact may overrepresent people with complaints. Record that limitation before proposing a wider pilot. The output is a revisable decision, not a certificate that an idea has been validated.