APP-HCD-01 - Constructed ImpactReview-17 decision

ImpactReview-17 is a constructed pedagogical application. Its answer key, ImpactReview-17-Domain-Oracle-v1, supplies engineering correctness and criticality criteria for PumpWorks-Config-Baseline-42 and AI-Assist-M4. Its observations and decision conditions are stipulated so that the reader can follow demand derivation, diagnosis, and profile comparison. Operational use needs qualified criteria and evidence for the actual person and Work.

The exact human holder is Engineer-P14. The later Work is independent review of AI-proposed change-impact claims for mixed physical/software pump-controller changes before separate integration, assurance, and release decisions. The constructed 2026-Q4 conditions include current sources and configuration, a forty-five-minute case limit, the named AI assistant, recoverable audit records, and bounded safety and maintenance specialist returns. Final integration, safety, release, and employment authority remain elsewhere. A routine reminder to construct and test an alternative configuration hypothesis is allowed; the engineer must supply the case-specific rival and its test. A cue giving that rival would no longer test the required human contribution.

HCD.1 uses eight stipulated representative review occurrences—four familiar and four unfamiliar, four with plausible wrong AI dependency claims, two needing specialist return—to return qualified demand only inside the constructed case. Engineer-P14 must recognize applicability, trace claims to source and configuration, independently challenge plausible AI output, delimit specialist results, return auditable evidence, and detect every seeded safety-critical omission under the protected conditions.

The constructed baseline has correct trace and evidence return in 8/8 cases, correct specialist routing in 2/2 relevant cases, and rejection of 2/4 plausible wrong AI claims. In the two unfamiliar misses, a neutral E.23.CAE-shaped probe preserves Method, source, tool, time, assignment, authority, and specialist route. Engineer-P14 recognizes the need for independent checking but does not generate and test a competing configuration hypothesis. HCD.3 returns the qualified target generate and test an independent configuration counter-hypothesis before accepting plausible AI output in unfamiliar variants, with the small sample and live prompt-design rival explicit.

HCD.4 keeps four rows:

Engineer-P14 is the person-level DecisionSubject for one self-development-focus choice inside a separately pre-authorized twelve-hour allocation. The pre-existing options are: O1 preserve without selecting change; O2 prioritize independent AI-output challenge for later programme comparison while preserving the supported claims; and O3 defer a human priority and return the prompt-design question to the Method/tool owner.

The shared evidential comparison retains the small sample and prompt rival. Use two further unfamiliar seeded-error cases, one first attempt per case: a neutral prompt on one, an explicit reminder of the operation on the other, using the prompt definitions in HCD.3:5.1. Success only with that reminder can support performance under the permitted assistance and favour returning prompt design to its owner; it leaves self-initiated recall open. Keep the same sources, tools, specialist routes, and forty-five-minute limit per case. Up to ninety minutes for attempts and thirty minutes for preparation and comparison fit the two-hour probe budget inside the twelve-hour allocation. The observations can still change the O2-versus-O3 comparison, but differences between the cases remain unresolved; this is not an isolated causal prompt test. The probe is feasible and worth its cost, so the explicit C.11 result is ChoiceResult = probe again. No profile option or programme is selected. The separately named integration, safety, release, and employment decisions remain with their authorized owners.