Link to current text
B.3:8 - Common anti-patterns and repairs
| Anti-pattern | Why it fails | Repair |
| Universal weakest-link formula | min can be an upper bound for a conjunction and can ignore alternatives or dependence. | Use the exact dependency model and domain rule; otherwise report inputs separately. |
| Ordinal penalty subtraction | An ordinal label is subtracted from a probability or ratio quantity. | Use a calibrated mapping to the same quantity when one exists, or keep the values separate. |
| Formality raises assurance | A more formal wrong model receives a better result. | State the inspectability gain and the uncertainty it closes; keep truth and empirical adequacy separate. |
| Same letter, different property | R names system reliability in one row and evidence quality in another. | Give each characteristic its own bearer and definition. |
| Safety-case inflation | A warning, access value, or consequential display triggers a generic B.3 package. | Use the direct domain pattern; apply B.3 only to an actual assurance claim. |
| Evidence creates truth | New evidence is said to make the target fact obtain. | Revise the warrant or disposition unless the world-side facts changed. |
| Assessment-record collapse | A checklist, trace, witness, or note is treated as the assessment Work or result. | Identify each object separately and add only what the use consumes. |
| Design/run chimera | Blueprint evidence and runtime observations are merged into one score. | Produce separate results and compare them. |