HCD.3:5.1 - Constructed Positive Case: ImpactReview-17

ImpactReview-17 is a constructed pedagogical case. ImpactReview-17-Domain-Oracle-v1 supplies engineering correctness and criticality criteria for PumpWorks-Config-Baseline-42 and AI-Assist-M4. The baseline observations and neutral-probe results below are stipulated case data.

For human holder Engineer-P14, the constructed HCD.1 account demands independent review of AI-proposed change-impact claims under a forty-five-minute limit with current sources, configuration trace, auditable evidence, and bounded safety and maintenance specialist returns. The baseline stipulates correct source/configuration trace and evidence return in eight of eight cases, correct specialist routing in two of two relevant cases, and rejection of two of four plausible wrong AI claims. Both misses occur in unfamiliar variants and are labelled safety-critical only by the constructed oracle.

Several rivals would change the next Work. The applicable MethodDescription, source access, tool configuration, time, assignment, authority, and specialist route are present in the case, so those branches do not explain the two misses. An E.23.CAE-shaped neutral probe preserves those conditions. Engineer-P14 recognizes that independent checking is required but does not generate and test a competing configuration hypothesis on either unfamiliar miss.

HCD.3 therefore returns this qualified intervention target:

Generate and test an independent configuration counter-hypothesis before accepting plausible AI output in unfamiliar variants.

The stipulated review permits an ordinary reminder to construct and test a competing hypothesis, but the engineer must supply the case-specific rival and its test. In the proposed comparison, the neutral prompt is “Review this AI claim against the current configuration.” The other prompt adds “Before accepting it, construct and test a competing configuration hypothesis.” Neither supplies the alternative or tells the engineer which claim to reject.

The evidence is narrow: two unfamiliar misses inside eight constructed cases. A prompt-design rival remains live because the neutral prompt may have suppressed the counter-hypothesis move. The protected conditions are the same source, configuration, AI assistant, time limit, authority, and specialist routes. The intended observation is whether Engineer-P14 produces and tests a competing configuration hypothesis and rejects the planted wrong claim in each of two further unfamiliar cases: one first attempt with a neutral prompt, and one first attempt on the other case with an explicit counter-hypothesis prompt. Success with the reminder supports performance under that permitted condition and may justify returning prompt design to the Method/tool owner; it does not establish self-initiated use of the operation. Different cases leave case-difference uncertainty; the comparison informs the next decision without isolating a causal prompt effect. Reopen when the prompt, configuration, case family, source access, tool posture, time, observation count, or target-domain oracle changes.

The result is a bounded target for a later development decision, not an assignment to practice. Operational use needs qualified current domain criteria and observations for the actual person and Work, including evidence that distinguishes the live rivals.