| task-set omission bias | One visible task defines the arrangement while another mandatory use remains invisible. | Build one row per named user and use, then compare every candidate against the same bounded set. |
| forced-winner bias | A local minimum or one observed pass becomes the single arrangement despite incomparability. | Preserve retained candidates, missing priorities, abstention, or a stated split boundary. |
| evidence-status bias | No test evidence is reported as failure or insufficiency. | Separate observed pass, observed failure, untested, feature-level insufficiency, and missing information. |
| repository bias | Uploading files is reported as a usable Method Base. | Test the declared task rows and retain the collection predicates separately. |
| platform bias | Feature breadth replaces the failed task set. | Include an element only for a named criterion, stop, cross-task dependency, or directly supported relation. |
| tool-authority bias | A recommendation or branch display becomes a decision or permission. | Keep tool interaction, user Work, permission, authority, and result separate. |
| structure-selection bias | A complete proposal table becomes an A.22 selected structure. | Require the selecting System, Method, Work, and participation or binding facts. |
| success inflation | One task pass becomes capability or Method-effectiveness evidence. | Bound the result to observed Work and return neighboring claims separately. |