Evidence Gaps and Research Opportunities
An evidence gap is a missing step between a supported result and a desired conclusion. It can suggest a strong research project when the missing step is consequential, accessible, and not already resolved elsewhere. It can also reveal that a proposed conclusion cannot be tested with the available method.
Classify what is missing
Section titled “Classify what is missing”| Gap | A useful contribution | What would still remain undecided |
|---|---|---|
| Reproduction | Recover a published calculation or analysis with documented artifacts | Whether the physical model describes the experiment |
| Independent confirmation | Test the claim with new data or an independent method | Generality outside the tested regime |
| Identification | Find an observable that separates competing mechanisms | The behavior of mechanisms not included in the comparison |
| Scaling | Control size, time, precision, or resource dependence | Extrapolation beyond the controlled range |
| Integration | Demonstrate components operating together under matched conditions | Reliability at a larger delivered workload |
| Operational definition | Specify and implement the measured quantity or allowed transformation | Equivalence to other definitions under different assumptions |
The categories overlap, but naming the dominant gap prevents a project from treating one kind of success as evidence for another. Evidence standards gives the corresponding claim contracts.
Work backward from the desired conclusion
Section titled “Work backward from the desired conclusion”Suppose a memory experiment shows lower logical error as code size grows. A useful computational claim additionally needs protected operations, resource production, routing, decoding, and a verified workload. This does not diminish the memory result. It identifies integration gaps that should be assessed separately in fault-tolerant computing.
Suppose a numerical study finds a long-lived plateau. A phase claim needs control of the relevant limits and competing explanations. The feasible project may be a new diagnostic or a bound on what current sizes can resolve, rather than a larger version of the same plot.
Choose a gap you can actually reduce
Section titled “Choose a gap you can actually reduce”A graduate entry project should have an accessible baseline, a single discriminating extension, and a useful fallback. For a numerical gap, consult convergence studies and separate solver accuracy from physical approximation. For an experimental gap, confirm instrument access and sensitivity before proposing a new apparatus. For a proof gap, restrict the statement enough that a counterexample or partial theorem is attainable.
Ask whether closing the gap changes a scientific decision. Reproducing a known figure may be essential preparation but is not automatically a new contribution. A carefully designed negative result can be more informative than a small positive improvement if it rules out a plausible mechanism or exposes a hidden cost.
A project-design exercise
Section titled “A project-design exercise”Select two competing explanations from a topic page. Write their predictions for the same observable under identical assumptions. Identify a parameter region where the predictions separate and estimate the uncertainty needed to discriminate them. Then check whether the required regime is physically or computationally accessible.
Deliver a comparison plot or proof outline, an error budget, and a stopping criterion. If no accessible observable separates the models, report that identifiability limit rather than manufacture a preference. A research extension might propose a new intervention that removes the ambiguity.
When the gap closes
Section titled “When the gap closes”Update the assessment when a result addresses the exact missing step. A new paper on the same topic may leave it untouched; a technical correction may change it completely. Preserve the source version and the reason for the revised conclusion. Source monitoring helps maintain that record without treating publication volume as evidence strength.