After a poison tabletop, the assessor office adds trustworthiness metrics to the valuation model scorecard beside accuracy. What lifecycle practice is that?
Select an answer to reveal the explanation.
Short Explanation
Accuracy alone is a shiny speedometer—trustworthiness checks whether the needle can be tricked.
Full Explanation
Adding trustworthiness metrics to the valuation model scorecard beside accuracy after a poison tabletop is the lifecycle practice of measuring trustworthiness alongside accuracy for civic model acceptance. Accuracy on a clean test set can look strong even when the assessor model is fragile to poisoned parcels or manipulated features. Trustworthiness metrics—poison-test pass rates, provenance coverage, integrity-check success, stability under adversarial feature noise—tell the municipality whether acceptance should proceed. Institutionalizing both prevents shipping solely because F1 looks good after a poison-focused exercise.
Archiving prior assessment model versions is a rollback recovery task, not a trustworthiness metric for model acceptance. Documenting comparable sales lineage supports appraisal audit reproducibility, but it does not quantify model trustworthiness or acceptance readiness. Calibrating mass appraisal coefficients for statutory ratio study compliance is a statutory modeling requirement, not a trustworthiness gate for accepting a civic model. Distractors fail because they address backup, audit, or compliance tasks rather than acceptance criteria for civic ML.
Exam caveat: trustworthiness metrics complement—not replace—accuracy, fairness, and operational monitoring. Define metrics that are testable and tied to the poison scenarios exercised; vague trust scores without methods fail audits. Avoid gaming: keep poison tests out of the training loop unless intentionally used for adversarial training under policy.
Operational check: update the assessor model scorecard with explicit gates (for example, minimum pass rate on parcel-feature poison suites and full digest verification on training extracts). Block promotion when trustworthiness gates fail even if accuracy passes. Review scorecard evidence in change control with assessor IT and security before production valuation use.