What Is the Reproducibility Crisis?
A Philosophical Perspective

Reproducibility in Science

Reproducibility refers to the capacity for the same result to be obtained under the same conditions. In science, the ability of independent researchers to confirm a finding is widely regarded as a cornerstone of objective knowledge.

In recent decades, however, a reproducibility crisis has been identified across psychology, medicine, biology, and related fields. A significant proportion of published findings have failed to replicate when independently tested.

Standard Explanations for Reproducibility Failure

Existing accounts of replication failure typically appeal to factors such as underpowered samples, selective reporting, p-hacking, methodological inconsistency, and researcher degrees of freedom. These factors are genuine and important.

But each of them rests on a shared assumption that is rarely made explicit.

This presupposition is not trivial. It is typically expressed in natural language, which lacks the resources to determine whether two problem instances are structurally identical.

Reproducibility Failure and Problem Identity

Before asking whether a result has been reproduced, a prior question must be answered: are the two studies actually addressing the same problem?

If the problems differ structurally, a divergence in results is to be expected. In that case, what appears to be a failure of reproducibility is, more precisely, a failure of problem identity.

The SRF Research Programme argues that many instances of the reproducibility crisis fall into this category — and that diagnosing them correctly requires a formal account of what problem identity means.

Formalising Problem Identity

To ask rigorously whether two problems are the same, a problem must be understood not merely as a linguistic description, but as a structured object.

The SRF defines a scientific problem as a triple:

Here S is the state space (the set of distinguishable data states), τ is the transition structure (the dynamic relations between states), and Γ is the constraint set (the boundary conditions that individuate the problem). Two problems are identical if and only if there exists a structural isomorphism between them — a bijection preserving all three components.

The DTC Conditions

The isomorphism condition yields three independently necessary preservation requirements.

• D

  Data Preservation

  Whether the two investigations operate on the same kind of data. The structure of the state space must correspond across both problems.

• T

  Transition Preservation

  Whether the relational and dynamic structure between data states is preserved. The logic of transitions must be maintained across both problems.

• C

  Constraint Preservation

  Whether the boundary conditions and admissibility criteria that constitute the problem as such are held constant across both instances.

When all three DTC conditions are jointly satisfied, the problems are structurally identical, and reproducibility evaluation is well-founded. When any condition fails, the comparison is conceptually underdetermined.

A Reframing of the Crisis

This framework does not replace the existing explanations of replication failure. It operates at a prior level: it asks whether the comparison itself is legitimate before statistical or methodological factors are considered.

The reproducibility crisis, on this view, is not only a technical problem. It is, in part, a conceptual one — a failure to ask what it means for two investigations to address the same problem.