| “Research proves the architecture” | Connect the source claim to the project claim, transfer limits, evidence use, and decision effect; keep the architecture decision separate. |
| “The model is validated” | State the modeled subject, purpose, configuration, comparison basis, result, uncertainty, unsupported use, and currentness. |
| Treat simulation as physical evidence | Keep model behavior separate from observations of the actual configuration; seek physical evidence only for claims that need it. |
| Treat test pass as acceptance or release | Separate test Work, criterion result, evidence use, acceptance rule, assurance, permission, and release decision. |
| Count reports or climb one universal test ladder | Examine independence, shared assumptions, claim match, and decision value; use C.11 or specialist experimental design for further Work. |
| Hide a failed or negative trial | Record the result, validity limits, and alternatives it changes. |
| Accept a benchmark or AI confidence as general capability | State the task, population, configuration, Method, measure, result, checking Work, and transfer boundary. |
| Demand certainty before a reversible action | Use a stated reliance limit, accepted uncertainty, reversible choice, observation, and reopen condition. |