| favorite-method bias | The practitioner’s familiar design defines the question and rivals. | Hold the brief fixed and compare actual answer-changing relations with the relevant no-new-study outcome. |
| hierarchy bias | One design family is presumed strongest regardless of claim and threat. | Compare the specific evidence relation and criticism each option enables. |
| novelty-of-tool bias | A new sensor, model, AI workflow, or platform is treated as an informative design. | State the answer-changing contrast, threats, and validation relation independently of the tool. |
| false-stop bias | Constraint, fatigue, low priority, or desired certainty is called evidence sufficiency. | Tie EvidenceSufficientStop to the brief’s answer condition and qualified current evidence; an allowed uncertainty return does not meet that condition. |
| provisional-design bias | A selected label is retained while a load-bearing value is “TBD.” | Run the counterfactual; if the family or criticism can change, return ResearchDesignHold. |
| hold-everything bias | Harmless record, naming, or schedule details prevent selection. | Bound admissible values and prove that the family and criticism remain invariant. |
| reporting-as-design bias | A reporting guideline or protocol template is treated as design evidence. | Use it later for transparency where applicable; select the question-to-criticism relation here. |
| mixed-method accumulation | More methods are assumed to triangulate or strengthen automatically. | Give every component an irreducible evidence role and specify integration, disagreement, and partial-failure logic. |
| information-gain theater | A number is assigned without a defensible outcome model or decision use. | State which live uncertainty can change and how; use qualitative bounds when they are the honest resolution. |
| authority leakage | Selection is treated as ethics, safety, access, funding, or downstream-action approval. | Consume exact owner results and return the decision without transferring authority. |