A startup score can organise research, but it cannot express the probability of success unless it is calibrated against real outcomes. Here is how to use scored assessments without being misled.
A viability score feels decisive. It compresses a messy market, uncertain buyer behaviour and dozens of assumptions into one number. That convenience is also the danger.
A score can be a useful checklist summary. It is not automatically a probability, a forecast or proof that one idea is better than another.
Most early-stage evidence is incomplete. Search data can show interest without purchase intent. Reviews reveal pain among existing users but may not represent the wider market. Competitor revenue estimates can be wrong. Interviews contain memory and courtesy bias.
Turning these inputs into 73/100 does not remove those limitations. It hides them behind arithmetic.
The key question is not whether a tool uses a score. It is what the score means, how it was constructed and whether it has been tested against the outcome people infer from it.
A scoring model may consistently rank one research brief above another. That can help triage opportunities. But a ranking is not a statement that an idea has a 73% chance of success.
To support a predictive claim, a provider would need:
Without that evidence, the number is an internal assessment scale. It may be thoughtfully designed, but it should not be read as an observed success rate.
A score typically combines market size, pain, competition, differentiation, timing and execution risk. The model must decide how much each factor matters.
Those weights may be reasonable, but they are still choices. They can also interact in ways a simple weighted average misses. A large market does not repair destructive unit economics. Strong user pain does not guarantee budget. A distinctive product does not remove procurement friction.
Ask whether a high total can average away a fatal weakness. In many businesses, one constraint—regulation, gross margin, access to distribution or an impossible integration—can dominate the result.
A founder shown 69 may feel compelled to improve the pitch until it becomes 72. That can turn research into score optimisation.
The better response is to inspect what would change the decision:
A useful assessment should make these questions easier to answer, even if the headline score disappears.
A provider may publish the distribution of scores across thousands of reports. This reveals how its model behaves: the median, spread and proportion above a threshold.
That is valuable transparency. It still does not show whether high-scoring companies succeeded more often than low-scoring companies. It is a benchmark of model outputs, not necessarily a benchmark of business outcomes.
Always read the definition beside the number.
Instead of asking whether an idea is 78/100 viable, evaluate the important claims separately:
Use the free Evidence Quality Scorecard to rate an important source for recency, independence, behaviour, commercial relevance and corroboration.
A score can help when it is used as:
Label it accurately. Keep the underlying dimensions visible. Do not imply a probability unless it has been calibrated as one.
Replace “this idea scores 78” with:
Current evidence supports a frequent problem among a defined customer group, but buyer urgency and switching economics remain unproven. Run three buyer interviews and one paid pilot proposal before building the integration.
That statement is less exciting. It is also more actionable and honest.
SignalCraft follows this evidence-led approach: observations, interpretation, contradictions and gaps remain separate. Read the methodology, inspect the claims register, or compare the approach with Preuve AI.