| training-default bias | A poor result becomes a course or practice assignment. | Compare decision-changing rivals and return the smallest supported target or a non-training result. |
| dispositional bias | A configuration-dependent observation becomes a stable trait, attitude, or motivation label. | Bind holder, Work, conditions, horizon, observations, and any still-plausible rivals. |
| tool-blame or human-blame bias | Responsibility is assigned to the AI tool or person before the joint arrangement is separated. | Vary protected conditions and keep human, tool, Method, prompt, authority, and specialist contributions distinct. |
| familiar-mechanism bias | “Forgot,” “misconception,” or “not motivated” substitutes for a controlled differential. | Use E.23.CAE where applicable and retain any still-plausible explanation that could change the next decision. |
| target-rhetoric bias | Naming autonomy, confidence, motivation, or practice is treated as evidence that it is limiting. | Require holder-specific evidence for the proposed target. A later claim that an intervention worked through that target also needs target-engagement evidence. |
| clinical appropriation bias | A disorder or treatment process is recast as development. | Return diagnosis, treatment, provider, safety, consent, and clinical outcomes to qualified clinical ownership. |
| surveillance bias | More data collection is assumed to be harmless and authorized. | Name declared use, participation, access, privacy or worker-governance conditions, and direct authority before collection. |
| intervention leakage | The target wording quietly names a programme, provider, schedule, or method. | Name the person’s difficulty with the required contribution, the smallest supported human change to target, and the intended observation; select any intervention separately. |