SourceFitnessTermAsFPFObject | “Fitness function” is written as the object under work. | Rewrite as ArchitectureCharacteristicEvalProgram@Project and name evaluated rows, candidates, parity frame, eval operations, and receiving use. |
EvalAsCriterion | A benchmark, monitor, or test is named as the architecture characteristic. | Return to ACS; name the criterion, bearer, scale, proxy risk, and protected counter-characteristics before writing the eval. |
TestModeAsEvalWhole | The team only asks whether one candidate passes while the work question is variant comparison. | Keep the test for the hard constraint, then add eval result forms that compare candidates or expose the trade-off front. |
UnfairComparison | Candidates are compared under different budgets, evidence windows, environments, or missing-data rules. | Rebuild the parity frame or record the result as unusable for selection. |
ResultAsDecision | A rank, score, pass, or dashboard reading selects the architecture. | Treat the result as source material for an A.10 evidence-provenance account when an evidence claim is current, or as comparison input when comparison is current. Use A.19.CPM for explicit comparison, A.19.SelectorMechanism for set-returning selection, C.11 for local choice, G.5 for selected-set result declaration, and C.32.PAD for a project architecture decision. For publication, use E.17 for a source-backed face and source return and E.24.PUB for the publication occurrence and audience availability. |
SingleIndicatorGoodhart | Work improves one optimized indicator while an unmeasured architecture concern worsens. | Limit optimized indicators, add protected counter-characteristics, and open E.13 when proxy-to-value drift appears. |
LosingVariantAsError | A candidate that lost a planned eval is recorded as a mistake. | Record it as a variant result unless an expectation caused unplanned rework; keep useful learning in the variant archive. |
ProgramAsRunOrResult | The ACE record is said to execute, measure, evaluate, or produce the result, or one generic resultRef hides the result kind. | Recover each exact evaluator through A.13 and let A.15.1 independently admit the evaluation U.Work. Add an attribution ref only when the record expressly represents precise assignment-bound attribution; missing or failed F.6 leaves the Work intact. Add an operation application only when it independently obtains, and identify the typed result under the applicable pattern; keep the ACE record as the evaluation framing record. |